22145 - User contributions [en]

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-11-04T11:23:20Z

Carol: /* Practical Information regarding Exam */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AllergenicityPrediction Allergenicity Prediction]

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction Antibody Structure Prediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B Cell Epitope prediction]'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Practical Information regarding Exam ==

In order to get access to the written exam you should submit first all the exercises.

The exam has two parts:
:*A 2hs written exam on '''November 11th'''
:* An oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

:'''Regarding the Written exam:'''
: The exam will take place in '''Building 208, Holdlokale 3/065''' on November 11th.
: The exam starts at 8.45, you will have 2hs unless you have informed that you have been granted extra time.
: You need to bring your '''DTU Student card''' in order to get access to the exam.
: The '''exam will be provided in digital form, so you need to bring a computer''' with the required software installed.
: The exam does not allow the use of internet nor any means of communication with other students. Fail to comply with the [https://student.dtu.dk/en/exam/exam-cheating/dtu-code-of-honour code of honor] will be severely penalised.

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-11-04T11:23:06Z

Carol: /* Practical Information regarding Exam */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AllergenicityPrediction Allergenicity Prediction]

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction Antibody Structure Prediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B Cell Epitope prediction]'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Practical Information regarding Exam ==

In order to get access to the written exam you should submit first all the exercises.

The exam has two parts:
:*A 2hs written exam on '''November 11th'''
:* An oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

:Regarding the Written exam:
: The exam will take place in '''Building 208, Holdlokale 3/065''' on November 11th.
: The exam starts at 8.45, you will have 2hs unless you have informed that you have been granted extra time.
: You need to bring your '''DTU Student card''' in order to get access to the exam.
: The '''exam will be provided in digital form, so you need to bring a computer''' with the required software installed.
: The exam does not allow the use of internet nor any means of communication with other students. Fail to comply with the [https://student.dtu.dk/en/exam/exam-cheating/dtu-code-of-honour code of honor] will be severely penalised.

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-11-04T11:22:50Z

Carol: /* Practical Information regarding Exam */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AllergenicityPrediction Allergenicity Prediction]

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction Antibody Structure Prediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B Cell Epitope prediction]'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Practical Information regarding Exam ==

In order to get access to the written exam you should submit first all the exercises.

The exam has two parts:
:*A 2hs written exam on '''November 11th'''
:* An oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

:Regarding the Written exam:
The exam will take place in '''Building 208, Holdlokale 3/065''' on November 11th.
The exam starts at 8.45, you will have 2hs unless you have informed that you have been granted extra time.
You need to bring your '''DTU Student card''' in order to get access to the exam.
The '''exam will be provided in digital form, so you need to bring a computer''' with the required software installed.
The exam does not allow the use of internet nor any means of communication with other students. Fail to comply with the [https://student.dtu.dk/en/exam/exam-cheating/dtu-code-of-honour code of honor] will be severely penalised.

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-11-04T11:22:20Z

Carol: /* Exam */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AllergenicityPrediction Allergenicity Prediction]

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction Antibody Structure Prediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B Cell Epitope prediction]'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Practical Information regarding Exam ==

In order to get access to the written exam you should submit first all the exercises.

The exam has two parts:
:*A 2hs written exam on '''November 11th'''
:* An oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

Regarding the Written exam:
The exam will take place in '''Building 208, Holdlokale 3/065''' on November 11th.
The exam starts at 8.45, you will have 2hs unless you have informed that you have been granted extra time.
You need to bring your '''DTU Student card''' in order to get access to the exam.
The '''exam will be provided in digital form, so you need to bring a computer''' with the required software installed.
The exam does not allow the use of internet nor any means of communication with other students. Fail to comply with the [https://student.dtu.dk/en/exam/exam-cheating/dtu-code-of-honour code of honor] will be severely penalised.

Bcell Epitope prediction ans

2026-10-28T10:58:16Z

Carol: /* Q3. Which residues form the epitopes? Are they linear? Do they have a linear core? */

= B-cell epitope prediction =

== Analysis Tips and conventions ==

* '''Experimental epitope definition:''' An HA amino-acid residue belongs to the antibody-binding footprint if any of its non-hydrogen protein atoms is strictly less than 5.0 Å from a non-hydrogen protein atom of the selected antibody. Exclude waters and other non-protein molecules. Count a residue only once, even if several of its atoms contact the antibody.
* '''One interface at a time:''' For 7K39, use antigen chains A/B and one Fab (G/H); do not combine all three equivalent antibody copies.
* '''Residue numbering:''' The chain identifier and PDB residue number are part of the residue identity. For the supplied 7K39 BepiPred outputs, chain A sequence position = PDB residue number + 16; chain B sequence position = PDB residue number. This was confirmed by comparing the amino-acid sequences.
* '''Prediction evaluation:''' Predictions were compared on the 490 resolved amino acids in 7K39 HA chains A/B (318 in A and 172 in B). The default scores/flags were used unless stated otherwise.
* '''Important distinction:''' A residue outside one antibody's footprint is not necessarily a biologically impossible epitope; a different antibody could bind there. Thus “false positive” below means “outside the experimentally observed interface being tested.”

== Part A. Get your bearings in PyMOL (experimental epitopes) ==

=== Q1. List chain IDs for antigen versus antibody in each complex ===

{| class="wikitable"
! Experimental structure !! Antigen HA1 !! Antigen HA2 !! Antibody heavy chain !! Antibody light chain
|-
| '''4FQI''' (H5N1, CR9114) || A || B || H || L
|-
| '''7K39''' (H3N2, head-binding Fab) || A, C, E || B, D, F || H, J, L || G, I, K
|}

For the 7K39 calculations below use '''A/B''' as the antigen and '''G/H''' as a single Fab (G is light, H is heavy).

=== Q2. Show the epitopes, count residues, and describe secondary structure ===

Using the experimental PDB coordinates, excluding waters, and applying the 5 Å cutoff

{| class="wikitable"
! Complex !! HA1 (A) contact residues !! HA2 (B) contact residues !! Total !! Qualitative structural location
|-
| '''4FQI, stem-binding CR9114''' || 4 || 14 || '''18''' || Mainly at the HA stem, including an HA2 helical region and adjacent loops/turns.
|-
| '''7K39, head-binding Fab''' || 14 || 1 || '''15''' || Mainly on HA1 in the head, distributed among accessible loops and neighboring structural elements.
|}

'''Figure placeholders:''' Upload the two PyMOL images to MediaWiki before enabling the lines below.


To reproduce the selection in PyMOL, for 4FQI:
<pre>
fetch 4fqi
remove solvent
select Ag, (chain A+B)
select Fab, (chain H+L)
select epitope, byres (Fab around 5)
color red, epitope
</pre>
For 7K39, replace the Fab selection with chain G+H

=== Q3. Which residues form the epitopes? Are they linear? Do they have a linear core? ===

'''4FQI (18 contacts):'''
* HA1, '''A''': 291, 292, 293, 318.
* HA2, '''B''': 18, 19, 20, 21, 36, 37, 38, 41, 42, 45, 46, 48, 49, 52, 56.

'''7K39 (15 contacts):'''
* HA1, '''A''': 48, 49, 50, 90, 91, 92, 93, 269, 270, 271, 272, 273, 284, 285.
* HA2, '''B''': 69.

'''Interpretation:''' Both are predominantly '''conformational (discontinuous) epitopes''', because the contacting residues occur in sequence-distant regions that come together in three dimensions. They contain short contiguous stretches or potential ''linear cores'', for example 4FQI B:18–21 and B:36–38, and 7K39 A:90–93 and A:269–273. These patches do not make the overall epitopes linear or show that the corresponding isolated peptides would bind the antibody.

The antibody complexes do '''not''' target the same region: CR9114(4FQI) binds the '''stem''' (involving HA1 and HA2), while the 7K39 antibody binds primarily the '''head''' (mostly HA1). HA1=head and HA2=stem are useful teaching approximations, but HA1 also contributes to the structural stem.

== Part B. Sequence-based prediction: BepiPred-2.0 and BepiPred-3.0 ==

'''Data limitation:''' The provided BepiPred-2.0 and BepiPred-3.0 CSV files contain predictions for the '''7K39 H3N2 HA sequences'''. They were not run on the native '''4FQI H5N1''' HA sequences. The two structures contain homologous but non-identical antigens. The quantitative answers in Q4–Q5 therefore directly test the '''7K39 antibody footprint only'''. To evaluate 4FQI independently, run both predictors on the 4FQI FASTA sequences.

The supplied prediction scores were thresholded at '''0.50''' (BepiPred-2.0) and '''0.1512''' (BepiPred-3.0 default, unsmoothed).

=== Q4. Do BepiPred-2.0 predictions overlap the actual epitopes? ===

'''Yes for 7K39, but the result is not very selective.''' All '''15/15''' experimentally observed 7K39 residues are predicted as epitope, giving '''100% recall'''. However, '''407/490''' evaluated residues ('''83.1%''') are predicted as positive, of which 392 are outside the particular observed antibody footprint. Precision is only '''15/407 = 3.69%'''.

A BepiPred-2.0 score of 0.5 therefore recovers the experimental footprint by labeling nearly the whole antigen as a potential epitope. This is not good evidence that it can pinpoint where the specific 7K39 antibody binds. Its score-ranked '''average precision (AP) is 0.0306''', approximately the random-ranking baseline '''15/490 = 0.0306'''.

=== Q5. Does the BepiPred-3.0 prediction overlap now? ===

'''Only slightly for 7K39 at the default threshold.''' Using the ''unsmoothed'' BepiPred-3.0 output, '''2/15''' experimental contact residues are predicted ('''13.3% recall'''): '''A:92''' and '''A:285'''. Their corresponding BepiPred sequence positions are 108 and 301. There are '''128/490''' predicted positive residues (26.1% coverage); precision is '''2/128 = 1.56%''', and '''AP = 0.0297'''.

The optionally ''smoothed'' BepiPred-3.0 score detects '''0/15''' experimental contacts at the same numerical threshold and labels 88/490 residues positive. Smoothing may emphasize contiguous patches but is not guaranteed to improve overlap with a discontinuous antibody footprint.

'''Conclusion for Q4–Q5:''' BepiPred-2.0 has higher recall at its default threshold but very poor localization, while BepiPred-3.0 makes fewer predictions but misses most of this particular antibody footprint. Neither is demonstrably good at ranking the 7K39 contact residues.

=== Q6. What is different in BepiPred-2.0 compared with BepiPred-3.0? ===

{| class="wikitable"
! Feature !! BepiPred-2.0 !! BepiPred-3.0
|-
| Core method || Random Forest using sequence-derived features, trained with structural epitope annotations || Neural networks using protein language-model (ESM-2) representations, trained with updated structural epitope annotations
|-
| Input || Protein sequence || Protein sequence
|-
| Output focus || Mainly continuous/linear epitope prediction with smoothing || Residue-level prediction for potential linear and discontinuous epitope residues, with optional smoothing for linear epitope analysis
|-
| Default threshold used here || 0.50 || 0.1512
|}

BepiPred-3.0 captures richer sequence context using learned embeddings, but neither tool receives the identity or structure of the specific antibody. Predictions are thus potential general B-cell epitope residues, not antibody-specific docking poses. The BepiPred-3.0 structural training annotation uses an antibody-contact cutoff of 4 Å, whereas this exercise uses 5 Å; this difference can also affect observed overlap.

See [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0] and [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0].

== Part C. Structure-based prediction: DiscoTope-2.0 and DiscoTope-3.0 ==

'''Data:''' The uploaded ''Discotope2pred.txt'' and ''file_A_discotope3.csv'' / ''file_B_discotope3.csv'' correspond to the '''7K39 antigen A/B'''. At the supplied settings, DiscoTope-2.0 calls positive residues with combined score greater than approximately '''−3.7'''; DiscoTope-3.0 uses its '''calibrated score ≥ 0.90''' (moderate confidence). The numerical scales are different and cannot be compared directly.

'''Input caveat:''' The DiscoTope-2.0 output includes other HA chains and Fab chains, indicating that the uploaded structure was not strictly antigen-only. DiscoTope-3.0 outputs were reported separately for HA chains A and B. The comparison is therefore useful for the supplied results, but not a fully controlled benchmark of the methods.

=== Q7. Does DiscoTope-2.0 overlap with the actual epitope? ===

'''No, not at the selected threshold for 7K39.''' DiscoTope-2.0 classifies '''58/490''' antigen residues as epitopes (34 in HA1 chain A; 24 in HA2 chain B). '''None of the 15''' experimentally observed 7K39 contact residues is predicted positive, giving '''recall = 0%''' and '''precision = 0%'''. The ranking-based '''AP = 0.0274''', slightly below the 0.0306 random-ranking baseline.

These findings are specific to the supplied output and settings; the method should ideally be rerun on the identical antigen-only structure supplied to the newer version.

=== Q8. Compare the results obtained using DiscoTope-2.0 and DiscoTope-3.0 ===

{| class="wikitable"
! Measurement for 7K39 HA A/B !! DiscoTope-2.0 !! DiscoTope-3.0
|-
| Model concept || Spatial epitope propensities plus residue exposure and neighborhood information || Learned structural representations, including inverse-folding features and exposure information
|-
| Threshold used || Combined score > −3.7 || Calibrated score ≥ 0.90
|-
| Residues predicted positive || 58/490 || 77/490
|-
| Experimental contact residues recovered || 0/15 || '''4/15'''
|-
| Recall || 0% || '''26.7%'''
|-
| Precision || 0% || '''5.19%'''
|-
| Average precision (AP) || 0.0274 || '''0.0466'''
|}

DiscoTope-3.0 correctly identifies experimental contacts '''A:91, A:93, A:271 and B:69'''. At the supplied cutoffs, the two versions share only '''12 predicted-positive HA residues'''. DiscoTope-3.0 therefore agrees better with the known 7K39 footprint in the supplied results, although it still misses 11 of the 15 contacts and predicts 73 residues outside this one interface.

See [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0] and [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0].

=== Q9. Which prediction best matches the experimentally observed epitope? ===

To answer this, compare both '''recall''' and '''precision''', rather than declaring any overlapping prediction correct.

<pre>
Precision = TP / (TP + FP)
Recall = TP / (TP + FN)
</pre>

Here TP denotes predicted residues in the measured 5 Å antibody footprint; FP denotes predicted residues outside that ''specific'' footprint; and FN denotes measured contact residues missed by the method. A threshold-independent score-ranking summary is '''average precision (AP)''', calculated from a precision–recall curve. Its random-ranking baseline here is '''15/490 = 3.06%'''.

{| class="wikitable sortable"
! Method (7K39) !! Correct contacts !! All predicted positives !! Recall !! Precision !! AP !! ROC-AUC
|-
| BepiPred-2.0 || 15/15 || 407/490 || '''100%''' || 3.69% || 0.0306 || 0.492
|-
| BepiPred-3.0, unsmoothed || 2/15 || 128/490 || 13.3% || 1.56% || 0.0297 || 0.492
|-
| DiscoTope-2.0 || 0/15 || 58/490 || 0% || 0% || 0.0274 || 0.413
|-
| '''DiscoTope-3.0''' || '''4/15''' || '''77/490''' || 26.7% || '''5.19%''' || '''0.0466''' || '''0.626'''
|}

'''Best result for this experimental interface: DiscoTope-3.0''', if “best” means highest precision and AP in the supplied data. '''BepiPred-2.0 has the highest recall''', but almost no ability to discriminate the actual interface from the rest of HA at its default threshold. All AP values remain low in absolute terms. The table is not a general benchmark because only one measured antibody footprint has directly matched predictor outputs and the DiscoTope inputs were not identical.



== Part D. Predicting where a known antibody binds (AlphaFold) ==

=== Q10. Are there tools that predict where an isolated antibody binds to an antigen? What do the provided AlphaFold models show? ===

'''Yes.''' Given antigen and antibody sequences, complex-modeling methods such as '''AlphaFold 3''' can propose a three-dimensional antibody–antigen arrangement. Related approaches include '''Boltz''' and docking methods such as '''HADDOCK'''. Unlike BepiPred or DiscoTope, these methods use information about the ''specific antibody'' and aim to model its candidate binding pose. A predicted pose is a hypothesis, not proof of binding.

The provided ZIP archives contain five AlphaFold 3 models each for a hemagglutinin trimer plus antibody chains. They illustrate both successful and problematic complex prediction:

{| class="wikitable"
! Feature !! 4FQI AlphaFold 3 archive !! 7K39 AlphaFold 3 archive
|-
| Experimentally observed epitope || Stem || Head
|-
| Predicted binding location || '''Stem''', consistent across the five models || '''Near the stem''', not the experimental head site; the HA protomer contacted changes in some models
|-
| Example model 0 predicted 5 Å contact residues || 24 (8 in HA1 A, 16 in HA2 B) || 31 (12 in HA1 C, 19 in HA2 D)
|-
| Global complex ipTM, model 0 || 0.85 || 0.62
|-
| Antibody–HA chain-pair ipTM, across supplied models || Approximately '''0.49–0.70''' || Only '''0.08–0.13'''
|-
| Interpretation || Plausible binding pose consistent with the known stem location || Poorly supported antibody–antigen arrangement, despite a moderately high global ipTM value
|}

'''Crucial caution:''' Global ipTM reflects multiple interfaces in a trimeric complex, including HA–HA interfaces. For determining whether the antibody is positioned correctly, examine '''antibody–antigen pairwise ipTM''', interchain predicted aligned error (PAE), and the actual molecular contact geometry. In 7K39, the model-0 minimum antibody–HA pairwise PAE values are about 20–25 Å, indicating considerable uncertainty in the relative placement. This explains why an apparently folded antibody and HA trimer can nevertheless be docked incorrectly.

'''Conclusion:''' The AlphaFold 3 models offer a binding-site hypothesis, and the 4FQI prediction agrees qualitatively with experiment. The 7K39 prediction fails to place the antibody at its experimental head site and has very low antibody–antigen interface confidence. These models should be evaluated by comparison with experimental structures and, where appropriate, binding assays, mutagenesis or other validation.

For confidence-score interpretation see [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 output documentation].

== Overall answer to the exercise ==

The known 4FQI stem and 7K39 head epitopes are spatially discontinuous and are associated with different antibody-binding regions of HA. Sequence-based BepiPred tools can recover candidate epitope residues but, for the supplied 7K39 predictions, do not discriminate the measured antibody footprint well. Of the four submitted predictions, DiscoTope-3.0 gives the strongest agreement with that particular 7K39 footprint (AP 0.0466 versus a random baseline of 0.0306), though the result is still modest. When the identity of an antibody is known, complex prediction can offer a more specific binding pose, but the 7K39 AlphaFold example demonstrates the risk of incorrect docking and the importance of interface confidence metrics. Predictions should be treated as testable hypotheses rather than definitive experimental epitopes.

== Limitations and reproducibility ==

* All experimental epitope counts here use a strict 5 Å '''protein-heavy-atom''' cutoff without solvent; 4FQI = '''23''' and 7K39 = '''15'''.
* BepiPred and DiscoTope numerical performance is directly measurable from the uploaded outputs for '''7K39 only'''. Native-H5 predictions for 4FQI were not supplied, so a proper 4FQI precision–recall evaluation cannot be completed using these outputs.
* The BepiPred residue positions must be translated to PDB numbering (7K39 A: minus 16; 7K39 B: unchanged).
* The DiscoTope-2.0 input includes more chains than the DiscoTope-3.0 antigen-only result, so their numerical comparison is potentially confounded.
* AP is a precision–recall ranking summary; it is not the same quantity as ROC-AUC. The positive class is only 15/490 residues, so conventional classification accuracy is not a useful primary metric.
* Results against only one experimental antibody interface cannot establish that other exposed residues are never epitopes for other antibodies.

== Sources and software ==

* [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction DTU 22145 exercise]
* [https://www.rcsb.org/structure/4FQI PDB 4FQI — H5N1 HA–CR9114]
* [https://www.rcsb.org/structure/7K39 PDB 7K39 — H3N2 HA–head-binding antibody]
* [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0]
* [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0]
* [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 confidence and output documentation]

Bcell Epitope prediction ans

2026-10-28T10:56:45Z

Carol: /* Q2. Show the epitopes, count residues, and describe secondary structure */

= B-cell epitope prediction =

== Analysis Tips and conventions ==

* '''Experimental epitope definition:''' An HA amino-acid residue belongs to the antibody-binding footprint if any of its non-hydrogen protein atoms is strictly less than 5.0 Å from a non-hydrogen protein atom of the selected antibody. Exclude waters and other non-protein molecules. Count a residue only once, even if several of its atoms contact the antibody.
* '''One interface at a time:''' For 7K39, use antigen chains A/B and one Fab (G/H); do not combine all three equivalent antibody copies.
* '''Residue numbering:''' The chain identifier and PDB residue number are part of the residue identity. For the supplied 7K39 BepiPred outputs, chain A sequence position = PDB residue number + 16; chain B sequence position = PDB residue number. This was confirmed by comparing the amino-acid sequences.
* '''Prediction evaluation:''' Predictions were compared on the 490 resolved amino acids in 7K39 HA chains A/B (318 in A and 172 in B). The default scores/flags were used unless stated otherwise.
* '''Important distinction:''' A residue outside one antibody's footprint is not necessarily a biologically impossible epitope; a different antibody could bind there. Thus “false positive” below means “outside the experimentally observed interface being tested.”

== Part A. Get your bearings in PyMOL (experimental epitopes) ==

=== Q1. List chain IDs for antigen versus antibody in each complex ===

{| class="wikitable"
! Experimental structure !! Antigen HA1 !! Antigen HA2 !! Antibody heavy chain !! Antibody light chain
|-
| '''4FQI''' (H5N1, CR9114) || A || B || H || L
|-
| '''7K39''' (H3N2, head-binding Fab) || A, C, E || B, D, F || H, J, L || G, I, K
|}

For the 7K39 calculations below use '''A/B''' as the antigen and '''G/H''' as a single Fab (G is light, H is heavy).

=== Q2. Show the epitopes, count residues, and describe secondary structure ===

Using the experimental PDB coordinates, excluding waters, and applying the 5 Å cutoff

{| class="wikitable"
! Complex !! HA1 (A) contact residues !! HA2 (B) contact residues !! Total !! Qualitative structural location
|-
| '''4FQI, stem-binding CR9114''' || 4 || 14 || '''18''' || Mainly at the HA stem, including an HA2 helical region and adjacent loops/turns.
|-
| '''7K39, head-binding Fab''' || 14 || 1 || '''15''' || Mainly on HA1 in the head, distributed among accessible loops and neighboring structural elements.
|}

'''Figure placeholders:''' Upload the two PyMOL images to MediaWiki before enabling the lines below.


To reproduce the selection in PyMOL, for 4FQI:
<pre>
fetch 4fqi
remove solvent
select Ag, (chain A+B)
select Fab, (chain H+L)
select epitope, byres (Fab around 5)
color red, epitope
</pre>
For 7K39, replace the Fab selection with chain G+H

=== Q3. Which residues form the epitopes? Are they linear? Do they have a linear core? ===

'''4FQI (23 contacts):'''
* HA1, '''A''': 38, 40, 41, 42, 291, 292, 293, 318.
* HA2, '''B''': 18, 19, 20, 21, 36, 37, 38, 41, 42, 45, 46, 48, 49, 52, 56.

'''7K39 (15 contacts):'''
* HA1, '''A''': 48, 49, 50, 90, 91, 92, 93, 269, 270, 271, 272, 273, 284, 285.
* HA2, '''B''': 69.

'''Interpretation:''' Both are predominantly '''conformational (discontinuous) epitopes''', because the contacting residues occur in sequence-distant regions that come together in three dimensions. They contain short contiguous stretches or potential ''linear cores'', for example 4FQI B:18–21 and B:36–38, and 7K39 A:90–93 and A:269–273. These patches do not make the overall epitopes linear or show that the corresponding isolated peptides would bind the antibody.

The complexes do '''not''' target the same region: CR9114 binds the '''stem''' (involving HA1 and HA2), while the 7K39 antibody binds primarily the '''head''' (mostly HA1). HA1=head and HA2=stem are useful teaching approximations, but HA1 also contributes to the structural stem.

== Part B. Sequence-based prediction: BepiPred-2.0 and BepiPred-3.0 ==

'''Data limitation:''' The provided BepiPred-2.0 and BepiPred-3.0 CSV files contain predictions for the '''7K39 H3N2 HA sequences'''. They were not run on the native '''4FQI H5N1''' HA sequences. The two structures contain homologous but non-identical antigens. The quantitative answers in Q4–Q5 therefore directly test the '''7K39 antibody footprint only'''. To evaluate 4FQI independently, run both predictors on the 4FQI FASTA sequences.

The supplied prediction scores were thresholded at '''0.50''' (BepiPred-2.0) and '''0.1512''' (BepiPred-3.0 default, unsmoothed).

=== Q4. Do BepiPred-2.0 predictions overlap the actual epitopes? ===

'''Yes for 7K39, but the result is not very selective.''' All '''15/15''' experimentally observed 7K39 residues are predicted as epitope, giving '''100% recall'''. However, '''407/490''' evaluated residues ('''83.1%''') are predicted as positive, of which 392 are outside the particular observed antibody footprint. Precision is only '''15/407 = 3.69%'''.

A BepiPred-2.0 score of 0.5 therefore recovers the experimental footprint by labeling nearly the whole antigen as a potential epitope. This is not good evidence that it can pinpoint where the specific 7K39 antibody binds. Its score-ranked '''average precision (AP) is 0.0306''', approximately the random-ranking baseline '''15/490 = 0.0306'''.

=== Q5. Does the BepiPred-3.0 prediction overlap now? ===

'''Only slightly for 7K39 at the default threshold.''' Using the ''unsmoothed'' BepiPred-3.0 output, '''2/15''' experimental contact residues are predicted ('''13.3% recall'''): '''A:92''' and '''A:285'''. Their corresponding BepiPred sequence positions are 108 and 301. There are '''128/490''' predicted positive residues (26.1% coverage); precision is '''2/128 = 1.56%''', and '''AP = 0.0297'''.

The optionally ''smoothed'' BepiPred-3.0 score detects '''0/15''' experimental contacts at the same numerical threshold and labels 88/490 residues positive. Smoothing may emphasize contiguous patches but is not guaranteed to improve overlap with a discontinuous antibody footprint.

'''Conclusion for Q4–Q5:''' BepiPred-2.0 has higher recall at its default threshold but very poor localization, while BepiPred-3.0 makes fewer predictions but misses most of this particular antibody footprint. Neither is demonstrably good at ranking the 7K39 contact residues.

=== Q6. What is different in BepiPred-2.0 compared with BepiPred-3.0? ===

{| class="wikitable"
! Feature !! BepiPred-2.0 !! BepiPred-3.0
|-
| Core method || Random Forest using sequence-derived features, trained with structural epitope annotations || Neural networks using protein language-model (ESM-2) representations, trained with updated structural epitope annotations
|-
| Input || Protein sequence || Protein sequence
|-
| Output focus || Mainly continuous/linear epitope prediction with smoothing || Residue-level prediction for potential linear and discontinuous epitope residues, with optional smoothing for linear epitope analysis
|-
| Default threshold used here || 0.50 || 0.1512
|}

BepiPred-3.0 captures richer sequence context using learned embeddings, but neither tool receives the identity or structure of the specific antibody. Predictions are thus potential general B-cell epitope residues, not antibody-specific docking poses. The BepiPred-3.0 structural training annotation uses an antibody-contact cutoff of 4 Å, whereas this exercise uses 5 Å; this difference can also affect observed overlap.

See [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0] and [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0].

== Part C. Structure-based prediction: DiscoTope-2.0 and DiscoTope-3.0 ==

'''Data:''' The uploaded ''Discotope2pred.txt'' and ''file_A_discotope3.csv'' / ''file_B_discotope3.csv'' correspond to the '''7K39 antigen A/B'''. At the supplied settings, DiscoTope-2.0 calls positive residues with combined score greater than approximately '''−3.7'''; DiscoTope-3.0 uses its '''calibrated score ≥ 0.90''' (moderate confidence). The numerical scales are different and cannot be compared directly.

'''Input caveat:''' The DiscoTope-2.0 output includes other HA chains and Fab chains, indicating that the uploaded structure was not strictly antigen-only. DiscoTope-3.0 outputs were reported separately for HA chains A and B. The comparison is therefore useful for the supplied results, but not a fully controlled benchmark of the methods.

=== Q7. Does DiscoTope-2.0 overlap with the actual epitope? ===

'''No, not at the selected threshold for 7K39.''' DiscoTope-2.0 classifies '''58/490''' antigen residues as epitopes (34 in HA1 chain A; 24 in HA2 chain B). '''None of the 15''' experimentally observed 7K39 contact residues is predicted positive, giving '''recall = 0%''' and '''precision = 0%'''. The ranking-based '''AP = 0.0274''', slightly below the 0.0306 random-ranking baseline.

These findings are specific to the supplied output and settings; the method should ideally be rerun on the identical antigen-only structure supplied to the newer version.

=== Q8. Compare the results obtained using DiscoTope-2.0 and DiscoTope-3.0 ===

{| class="wikitable"
! Measurement for 7K39 HA A/B !! DiscoTope-2.0 !! DiscoTope-3.0
|-
| Model concept || Spatial epitope propensities plus residue exposure and neighborhood information || Learned structural representations, including inverse-folding features and exposure information
|-
| Threshold used || Combined score > −3.7 || Calibrated score ≥ 0.90
|-
| Residues predicted positive || 58/490 || 77/490
|-
| Experimental contact residues recovered || 0/15 || '''4/15'''
|-
| Recall || 0% || '''26.7%'''
|-
| Precision || 0% || '''5.19%'''
|-
| Average precision (AP) || 0.0274 || '''0.0466'''
|}

DiscoTope-3.0 correctly identifies experimental contacts '''A:91, A:93, A:271 and B:69'''. At the supplied cutoffs, the two versions share only '''12 predicted-positive HA residues'''. DiscoTope-3.0 therefore agrees better with the known 7K39 footprint in the supplied results, although it still misses 11 of the 15 contacts and predicts 73 residues outside this one interface.

See [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0] and [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0].

=== Q9. Which prediction best matches the experimentally observed epitope? ===

To answer this, compare both '''recall''' and '''precision''', rather than declaring any overlapping prediction correct.

<pre>
Precision = TP / (TP + FP)
Recall = TP / (TP + FN)
</pre>

Here TP denotes predicted residues in the measured 5 Å antibody footprint; FP denotes predicted residues outside that ''specific'' footprint; and FN denotes measured contact residues missed by the method. A threshold-independent score-ranking summary is '''average precision (AP)''', calculated from a precision–recall curve. Its random-ranking baseline here is '''15/490 = 3.06%'''.

{| class="wikitable sortable"
! Method (7K39) !! Correct contacts !! All predicted positives !! Recall !! Precision !! AP !! ROC-AUC
|-
| BepiPred-2.0 || 15/15 || 407/490 || '''100%''' || 3.69% || 0.0306 || 0.492
|-
| BepiPred-3.0, unsmoothed || 2/15 || 128/490 || 13.3% || 1.56% || 0.0297 || 0.492
|-
| DiscoTope-2.0 || 0/15 || 58/490 || 0% || 0% || 0.0274 || 0.413
|-
| '''DiscoTope-3.0''' || '''4/15''' || '''77/490''' || 26.7% || '''5.19%''' || '''0.0466''' || '''0.626'''
|}

'''Best result for this experimental interface: DiscoTope-3.0''', if “best” means highest precision and AP in the supplied data. '''BepiPred-2.0 has the highest recall''', but almost no ability to discriminate the actual interface from the rest of HA at its default threshold. All AP values remain low in absolute terms. The table is not a general benchmark because only one measured antibody footprint has directly matched predictor outputs and the DiscoTope inputs were not identical.



== Part D. Predicting where a known antibody binds (AlphaFold) ==

=== Q10. Are there tools that predict where an isolated antibody binds to an antigen? What do the provided AlphaFold models show? ===

'''Yes.''' Given antigen and antibody sequences, complex-modeling methods such as '''AlphaFold 3''' can propose a three-dimensional antibody–antigen arrangement. Related approaches include '''Boltz''' and docking methods such as '''HADDOCK'''. Unlike BepiPred or DiscoTope, these methods use information about the ''specific antibody'' and aim to model its candidate binding pose. A predicted pose is a hypothesis, not proof of binding.

The provided ZIP archives contain five AlphaFold 3 models each for a hemagglutinin trimer plus antibody chains. They illustrate both successful and problematic complex prediction:

{| class="wikitable"
! Feature !! 4FQI AlphaFold 3 archive !! 7K39 AlphaFold 3 archive
|-
| Experimentally observed epitope || Stem || Head
|-
| Predicted binding location || '''Stem''', consistent across the five models || '''Near the stem''', not the experimental head site; the HA protomer contacted changes in some models
|-
| Example model 0 predicted 5 Å contact residues || 24 (8 in HA1 A, 16 in HA2 B) || 31 (12 in HA1 C, 19 in HA2 D)
|-
| Global complex ipTM, model 0 || 0.85 || 0.62
|-
| Antibody–HA chain-pair ipTM, across supplied models || Approximately '''0.49–0.70''' || Only '''0.08–0.13'''
|-
| Interpretation || Plausible binding pose consistent with the known stem location || Poorly supported antibody–antigen arrangement, despite a moderately high global ipTM value
|}

'''Crucial caution:''' Global ipTM reflects multiple interfaces in a trimeric complex, including HA–HA interfaces. For determining whether the antibody is positioned correctly, examine '''antibody–antigen pairwise ipTM''', interchain predicted aligned error (PAE), and the actual molecular contact geometry. In 7K39, the model-0 minimum antibody–HA pairwise PAE values are about 20–25 Å, indicating considerable uncertainty in the relative placement. This explains why an apparently folded antibody and HA trimer can nevertheless be docked incorrectly.

'''Conclusion:''' The AlphaFold 3 models offer a binding-site hypothesis, and the 4FQI prediction agrees qualitatively with experiment. The 7K39 prediction fails to place the antibody at its experimental head site and has very low antibody–antigen interface confidence. These models should be evaluated by comparison with experimental structures and, where appropriate, binding assays, mutagenesis or other validation.

For confidence-score interpretation see [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 output documentation].

== Overall answer to the exercise ==

The known 4FQI stem and 7K39 head epitopes are spatially discontinuous and are associated with different antibody-binding regions of HA. Sequence-based BepiPred tools can recover candidate epitope residues but, for the supplied 7K39 predictions, do not discriminate the measured antibody footprint well. Of the four submitted predictions, DiscoTope-3.0 gives the strongest agreement with that particular 7K39 footprint (AP 0.0466 versus a random baseline of 0.0306), though the result is still modest. When the identity of an antibody is known, complex prediction can offer a more specific binding pose, but the 7K39 AlphaFold example demonstrates the risk of incorrect docking and the importance of interface confidence metrics. Predictions should be treated as testable hypotheses rather than definitive experimental epitopes.

== Limitations and reproducibility ==

* All experimental epitope counts here use a strict 5 Å '''protein-heavy-atom''' cutoff without solvent; 4FQI = '''23''' and 7K39 = '''15'''.
* BepiPred and DiscoTope numerical performance is directly measurable from the uploaded outputs for '''7K39 only'''. Native-H5 predictions for 4FQI were not supplied, so a proper 4FQI precision–recall evaluation cannot be completed using these outputs.
* The BepiPred residue positions must be translated to PDB numbering (7K39 A: minus 16; 7K39 B: unchanged).
* The DiscoTope-2.0 input includes more chains than the DiscoTope-3.0 antigen-only result, so their numerical comparison is potentially confounded.
* AP is a precision–recall ranking summary; it is not the same quantity as ROC-AUC. The positive class is only 15/490 residues, so conventional classification accuracy is not a useful primary metric.
* Results against only one experimental antibody interface cannot establish that other exposed residues are never epitopes for other antibodies.

== Sources and software ==

* [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction DTU 22145 exercise]
* [https://www.rcsb.org/structure/4FQI PDB 4FQI — H5N1 HA–CR9114]
* [https://www.rcsb.org/structure/7K39 PDB 7K39 — H3N2 HA–head-binding antibody]
* [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0]
* [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0]
* [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 confidence and output documentation]

Bcell Epitope prediction ans

2026-10-28T10:53:44Z

Carol: /* Q1. List chain IDs for antigen versus antibody in each complex */

= B-cell epitope prediction =

== Analysis Tips and conventions ==

* '''Experimental epitope definition:''' An HA amino-acid residue belongs to the antibody-binding footprint if any of its non-hydrogen protein atoms is strictly less than 5.0 Å from a non-hydrogen protein atom of the selected antibody. Exclude waters and other non-protein molecules. Count a residue only once, even if several of its atoms contact the antibody.
* '''One interface at a time:''' For 7K39, use antigen chains A/B and one Fab (G/H); do not combine all three equivalent antibody copies.
* '''Residue numbering:''' The chain identifier and PDB residue number are part of the residue identity. For the supplied 7K39 BepiPred outputs, chain A sequence position = PDB residue number + 16; chain B sequence position = PDB residue number. This was confirmed by comparing the amino-acid sequences.
* '''Prediction evaluation:''' Predictions were compared on the 490 resolved amino acids in 7K39 HA chains A/B (318 in A and 172 in B). The default scores/flags were used unless stated otherwise.
* '''Important distinction:''' A residue outside one antibody's footprint is not necessarily a biologically impossible epitope; a different antibody could bind there. Thus “false positive” below means “outside the experimentally observed interface being tested.”

== Part A. Get your bearings in PyMOL (experimental epitopes) ==

=== Q1. List chain IDs for antigen versus antibody in each complex ===

{| class="wikitable"
! Experimental structure !! Antigen HA1 !! Antigen HA2 !! Antibody heavy chain !! Antibody light chain
|-
| '''4FQI''' (H5N1, CR9114) || A || B || H || L
|-
| '''7K39''' (H3N2, head-binding Fab) || A, C, E || B, D, F || H, J, L || G, I, K
|}

For the 7K39 calculations below use '''A/B''' as the antigen and '''G/H''' as a single Fab (G is light, H is heavy).

=== Q2. Show the epitopes, count residues, and describe secondary structure ===

Using the experimental CIF coordinates, excluding waters, and applying the 5 Å protein-heavy-atom cutoff:

{| class="wikitable"
! Complex !! HA1 (A) contact residues !! HA2 (B) contact residues !! Total !! Qualitative structural location
|-
| '''4FQI, stem-binding CR9114''' || 8 || 15 || '''23''' || Mainly at the HA stem, including an HA2 helical region and adjacent loops/turns.
|-
| '''7K39, head-binding Fab''' || 14 || 1 || '''15''' || Mainly on HA1 in the head, distributed among accessible loops and neighboring structural elements.
|}

'''Figure placeholders:''' Upload the two PyMOL images to MediaWiki before enabling the lines below.


'''Why the previous 4FQI counts differed:''' The earlier PyMOL screenshot highlighted 18 residues (4 in A and 14 in B). A complete calculation with both antibody protein chains identifies another five within 5 Å: '''A:38, A:40, A:41, A:42 and B:52''', producing 23. An earlier estimate of 25 mistakenly allowed water atoms from the antibody-chain records to contribute contacts, adding A:18 and B:53. The protein-only, solvent-excluded result is '''23'''.

To reproduce the selection in PyMOL, for 4FQI:
<pre>
fetch 4fqi
remove solvent
select Ag, (chain A+B)
select Fab, (chain H+L)
select epitope, byres (Fab around 5)
color red, epitope
</pre>
For 7K39, replace the Fab selection with chain G+H

=== Q3. Which residues form the epitopes? Are they linear? Do they have a linear core? ===

'''4FQI (23 contacts):'''
* HA1, '''A''': 38, 40, 41, 42, 291, 292, 293, 318.
* HA2, '''B''': 18, 19, 20, 21, 36, 37, 38, 41, 42, 45, 46, 48, 49, 52, 56.

'''7K39 (15 contacts):'''
* HA1, '''A''': 48, 49, 50, 90, 91, 92, 93, 269, 270, 271, 272, 273, 284, 285.
* HA2, '''B''': 69.

'''Interpretation:''' Both are predominantly '''conformational (discontinuous) epitopes''', because the contacting residues occur in sequence-distant regions that come together in three dimensions. They contain short contiguous stretches or potential ''linear cores'', for example 4FQI B:18–21 and B:36–38, and 7K39 A:90–93 and A:269–273. These patches do not make the overall epitopes linear or show that the corresponding isolated peptides would bind the antibody.

The complexes do '''not''' target the same region: CR9114 binds the '''stem''' (involving HA1 and HA2), while the 7K39 antibody binds primarily the '''head''' (mostly HA1). HA1=head and HA2=stem are useful teaching approximations, but HA1 also contributes to the structural stem.

== Part B. Sequence-based prediction: BepiPred-2.0 and BepiPred-3.0 ==

'''Data limitation:''' The provided BepiPred-2.0 and BepiPred-3.0 CSV files contain predictions for the '''7K39 H3N2 HA sequences'''. They were not run on the native '''4FQI H5N1''' HA sequences. The two structures contain homologous but non-identical antigens. The quantitative answers in Q4–Q5 therefore directly test the '''7K39 antibody footprint only'''. To evaluate 4FQI independently, run both predictors on the 4FQI FASTA sequences.

The supplied prediction scores were thresholded at '''0.50''' (BepiPred-2.0) and '''0.1512''' (BepiPred-3.0 default, unsmoothed).

=== Q4. Do BepiPred-2.0 predictions overlap the actual epitopes? ===

'''Yes for 7K39, but the result is not very selective.''' All '''15/15''' experimentally observed 7K39 residues are predicted as epitope, giving '''100% recall'''. However, '''407/490''' evaluated residues ('''83.1%''') are predicted as positive, of which 392 are outside the particular observed antibody footprint. Precision is only '''15/407 = 3.69%'''.

A BepiPred-2.0 score of 0.5 therefore recovers the experimental footprint by labeling nearly the whole antigen as a potential epitope. This is not good evidence that it can pinpoint where the specific 7K39 antibody binds. Its score-ranked '''average precision (AP) is 0.0306''', approximately the random-ranking baseline '''15/490 = 0.0306'''.

=== Q5. Does the BepiPred-3.0 prediction overlap now? ===

'''Only slightly for 7K39 at the default threshold.''' Using the ''unsmoothed'' BepiPred-3.0 output, '''2/15''' experimental contact residues are predicted ('''13.3% recall'''): '''A:92''' and '''A:285'''. Their corresponding BepiPred sequence positions are 108 and 301. There are '''128/490''' predicted positive residues (26.1% coverage); precision is '''2/128 = 1.56%''', and '''AP = 0.0297'''.

The optionally ''smoothed'' BepiPred-3.0 score detects '''0/15''' experimental contacts at the same numerical threshold and labels 88/490 residues positive. Smoothing may emphasize contiguous patches but is not guaranteed to improve overlap with a discontinuous antibody footprint.

'''Conclusion for Q4–Q5:''' BepiPred-2.0 has higher recall at its default threshold but very poor localization, while BepiPred-3.0 makes fewer predictions but misses most of this particular antibody footprint. Neither is demonstrably good at ranking the 7K39 contact residues.

=== Q6. What is different in BepiPred-2.0 compared with BepiPred-3.0? ===

{| class="wikitable"
! Feature !! BepiPred-2.0 !! BepiPred-3.0
|-
| Core method || Random Forest using sequence-derived features, trained with structural epitope annotations || Neural networks using protein language-model (ESM-2) representations, trained with updated structural epitope annotations
|-
| Input || Protein sequence || Protein sequence
|-
| Output focus || Mainly continuous/linear epitope prediction with smoothing || Residue-level prediction for potential linear and discontinuous epitope residues, with optional smoothing for linear epitope analysis
|-
| Default threshold used here || 0.50 || 0.1512
|}

BepiPred-3.0 captures richer sequence context using learned embeddings, but neither tool receives the identity or structure of the specific antibody. Predictions are thus potential general B-cell epitope residues, not antibody-specific docking poses. The BepiPred-3.0 structural training annotation uses an antibody-contact cutoff of 4 Å, whereas this exercise uses 5 Å; this difference can also affect observed overlap.

See [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0] and [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0].

== Part C. Structure-based prediction: DiscoTope-2.0 and DiscoTope-3.0 ==

'''Data:''' The uploaded ''Discotope2pred.txt'' and ''file_A_discotope3.csv'' / ''file_B_discotope3.csv'' correspond to the '''7K39 antigen A/B'''. At the supplied settings, DiscoTope-2.0 calls positive residues with combined score greater than approximately '''−3.7'''; DiscoTope-3.0 uses its '''calibrated score ≥ 0.90''' (moderate confidence). The numerical scales are different and cannot be compared directly.

'''Input caveat:''' The DiscoTope-2.0 output includes other HA chains and Fab chains, indicating that the uploaded structure was not strictly antigen-only. DiscoTope-3.0 outputs were reported separately for HA chains A and B. The comparison is therefore useful for the supplied results, but not a fully controlled benchmark of the methods.

=== Q7. Does DiscoTope-2.0 overlap with the actual epitope? ===

'''No, not at the selected threshold for 7K39.''' DiscoTope-2.0 classifies '''58/490''' antigen residues as epitopes (34 in HA1 chain A; 24 in HA2 chain B). '''None of the 15''' experimentally observed 7K39 contact residues is predicted positive, giving '''recall = 0%''' and '''precision = 0%'''. The ranking-based '''AP = 0.0274''', slightly below the 0.0306 random-ranking baseline.

These findings are specific to the supplied output and settings; the method should ideally be rerun on the identical antigen-only structure supplied to the newer version.

=== Q8. Compare the results obtained using DiscoTope-2.0 and DiscoTope-3.0 ===

{| class="wikitable"
! Measurement for 7K39 HA A/B !! DiscoTope-2.0 !! DiscoTope-3.0
|-
| Model concept || Spatial epitope propensities plus residue exposure and neighborhood information || Learned structural representations, including inverse-folding features and exposure information
|-
| Threshold used || Combined score > −3.7 || Calibrated score ≥ 0.90
|-
| Residues predicted positive || 58/490 || 77/490
|-
| Experimental contact residues recovered || 0/15 || '''4/15'''
|-
| Recall || 0% || '''26.7%'''
|-
| Precision || 0% || '''5.19%'''
|-
| Average precision (AP) || 0.0274 || '''0.0466'''
|}

DiscoTope-3.0 correctly identifies experimental contacts '''A:91, A:93, A:271 and B:69'''. At the supplied cutoffs, the two versions share only '''12 predicted-positive HA residues'''. DiscoTope-3.0 therefore agrees better with the known 7K39 footprint in the supplied results, although it still misses 11 of the 15 contacts and predicts 73 residues outside this one interface.

See [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0] and [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0].

=== Q9. Which prediction best matches the experimentally observed epitope? ===

To answer this, compare both '''recall''' and '''precision''', rather than declaring any overlapping prediction correct.

<pre>
Precision = TP / (TP + FP)
Recall = TP / (TP + FN)
</pre>

Here TP denotes predicted residues in the measured 5 Å antibody footprint; FP denotes predicted residues outside that ''specific'' footprint; and FN denotes measured contact residues missed by the method. A threshold-independent score-ranking summary is '''average precision (AP)''', calculated from a precision–recall curve. Its random-ranking baseline here is '''15/490 = 3.06%'''.

{| class="wikitable sortable"
! Method (7K39) !! Correct contacts !! All predicted positives !! Recall !! Precision !! AP !! ROC-AUC
|-
| BepiPred-2.0 || 15/15 || 407/490 || '''100%''' || 3.69% || 0.0306 || 0.492
|-
| BepiPred-3.0, unsmoothed || 2/15 || 128/490 || 13.3% || 1.56% || 0.0297 || 0.492
|-
| DiscoTope-2.0 || 0/15 || 58/490 || 0% || 0% || 0.0274 || 0.413
|-
| '''DiscoTope-3.0''' || '''4/15''' || '''77/490''' || 26.7% || '''5.19%''' || '''0.0466''' || '''0.626'''
|}

'''Best result for this experimental interface: DiscoTope-3.0''', if “best” means highest precision and AP in the supplied data. '''BepiPred-2.0 has the highest recall''', but almost no ability to discriminate the actual interface from the rest of HA at its default threshold. All AP values remain low in absolute terms. The table is not a general benchmark because only one measured antibody footprint has directly matched predictor outputs and the DiscoTope inputs were not identical.



== Part D. Predicting where a known antibody binds (AlphaFold) ==

=== Q10. Are there tools that predict where an isolated antibody binds to an antigen? What do the provided AlphaFold models show? ===

'''Yes.''' Given antigen and antibody sequences, complex-modeling methods such as '''AlphaFold 3''' can propose a three-dimensional antibody–antigen arrangement. Related approaches include '''Boltz''' and docking methods such as '''HADDOCK'''. Unlike BepiPred or DiscoTope, these methods use information about the ''specific antibody'' and aim to model its candidate binding pose. A predicted pose is a hypothesis, not proof of binding.

The provided ZIP archives contain five AlphaFold 3 models each for a hemagglutinin trimer plus antibody chains. They illustrate both successful and problematic complex prediction:

{| class="wikitable"
! Feature !! 4FQI AlphaFold 3 archive !! 7K39 AlphaFold 3 archive
|-
| Experimentally observed epitope || Stem || Head
|-
| Predicted binding location || '''Stem''', consistent across the five models || '''Near the stem''', not the experimental head site; the HA protomer contacted changes in some models
|-
| Example model 0 predicted 5 Å contact residues || 24 (8 in HA1 A, 16 in HA2 B) || 31 (12 in HA1 C, 19 in HA2 D)
|-
| Global complex ipTM, model 0 || 0.85 || 0.62
|-
| Antibody–HA chain-pair ipTM, across supplied models || Approximately '''0.49–0.70''' || Only '''0.08–0.13'''
|-
| Interpretation || Plausible binding pose consistent with the known stem location || Poorly supported antibody–antigen arrangement, despite a moderately high global ipTM value
|}

'''Crucial caution:''' Global ipTM reflects multiple interfaces in a trimeric complex, including HA–HA interfaces. For determining whether the antibody is positioned correctly, examine '''antibody–antigen pairwise ipTM''', interchain predicted aligned error (PAE), and the actual molecular contact geometry. In 7K39, the model-0 minimum antibody–HA pairwise PAE values are about 20–25 Å, indicating considerable uncertainty in the relative placement. This explains why an apparently folded antibody and HA trimer can nevertheless be docked incorrectly.

'''Conclusion:''' The AlphaFold 3 models offer a binding-site hypothesis, and the 4FQI prediction agrees qualitatively with experiment. The 7K39 prediction fails to place the antibody at its experimental head site and has very low antibody–antigen interface confidence. These models should be evaluated by comparison with experimental structures and, where appropriate, binding assays, mutagenesis or other validation.

For confidence-score interpretation see [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 output documentation].

== Overall answer to the exercise ==

The known 4FQI stem and 7K39 head epitopes are spatially discontinuous and are associated with different antibody-binding regions of HA. Sequence-based BepiPred tools can recover candidate epitope residues but, for the supplied 7K39 predictions, do not discriminate the measured antibody footprint well. Of the four submitted predictions, DiscoTope-3.0 gives the strongest agreement with that particular 7K39 footprint (AP 0.0466 versus a random baseline of 0.0306), though the result is still modest. When the identity of an antibody is known, complex prediction can offer a more specific binding pose, but the 7K39 AlphaFold example demonstrates the risk of incorrect docking and the importance of interface confidence metrics. Predictions should be treated as testable hypotheses rather than definitive experimental epitopes.

== Limitations and reproducibility ==

* All experimental epitope counts here use a strict 5 Å '''protein-heavy-atom''' cutoff without solvent; 4FQI = '''23''' and 7K39 = '''15'''.
* BepiPred and DiscoTope numerical performance is directly measurable from the uploaded outputs for '''7K39 only'''. Native-H5 predictions for 4FQI were not supplied, so a proper 4FQI precision–recall evaluation cannot be completed using these outputs.
* The BepiPred residue positions must be translated to PDB numbering (7K39 A: minus 16; 7K39 B: unchanged).
* The DiscoTope-2.0 input includes more chains than the DiscoTope-3.0 antigen-only result, so their numerical comparison is potentially confounded.
* AP is a precision–recall ranking summary; it is not the same quantity as ROC-AUC. The positive class is only 15/490 residues, so conventional classification accuracy is not a useful primary metric.
* Results against only one experimental antibody interface cannot establish that other exposed residues are never epitopes for other antibodies.

== Sources and software ==

* [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction DTU 22145 exercise]
* [https://www.rcsb.org/structure/4FQI PDB 4FQI — H5N1 HA–CR9114]
* [https://www.rcsb.org/structure/7K39 PDB 7K39 — H3N2 HA–head-binding antibody]
* [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0]
* [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0]
* [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 confidence and output documentation]

Bcell Epitope prediction ans

2026-10-28T10:52:03Z

Carol: /* Q2. Show the epitopes, count residues, and describe secondary structure */

= B-cell epitope prediction =

== Analysis Tips and conventions ==

* '''Experimental epitope definition:''' An HA amino-acid residue belongs to the antibody-binding footprint if any of its non-hydrogen protein atoms is strictly less than 5.0 Å from a non-hydrogen protein atom of the selected antibody. Exclude waters and other non-protein molecules. Count a residue only once, even if several of its atoms contact the antibody.
* '''One interface at a time:''' For 7K39, use antigen chains A/B and one Fab (G/H); do not combine all three equivalent antibody copies.
* '''Residue numbering:''' The chain identifier and PDB residue number are part of the residue identity. For the supplied 7K39 BepiPred outputs, chain A sequence position = PDB residue number + 16; chain B sequence position = PDB residue number. This was confirmed by comparing the amino-acid sequences.
* '''Prediction evaluation:''' Predictions were compared on the 490 resolved amino acids in 7K39 HA chains A/B (318 in A and 172 in B). The default scores/flags were used unless stated otherwise.
* '''Important distinction:''' A residue outside one antibody's footprint is not necessarily a biologically impossible epitope; a different antibody could bind there. Thus “false positive” below means “outside the experimentally observed interface being tested.”

== Part A. Get your bearings in PyMOL (experimental epitopes) ==

=== Q1. List chain IDs for antigen versus antibody in each complex ===

{| class="wikitable"
! Experimental structure !! Antigen HA1 !! Antigen HA2 !! Antibody heavy chain !! Antibody light chain
|-
| '''4FQI''' (H5N1, CR9114) || A || B || H || L
|-
| '''7K39''' (H3N2, head-binding Fab) || A, C, E || B, D, F || H, J, L || G, I, K
|}

For the 7K39 calculations below use '''A/B''' as the antigen and '''G/H''' as a single Fab (G is light, H is heavy). The other equivalent antigen/Fab combinations are C/D with I/J, and E/F with K/L.

'''Note:''' The AlphaFold 3 ZIP files use their own chain labels. Do not transfer the experimental PDB chain mapping to the predicted complexes without checking their sequences.

=== Q2. Show the epitopes, count residues, and describe secondary structure ===

Using the experimental CIF coordinates, excluding waters, and applying the 5 Å protein-heavy-atom cutoff:

{| class="wikitable"
! Complex !! HA1 (A) contact residues !! HA2 (B) contact residues !! Total !! Qualitative structural location
|-
| '''4FQI, stem-binding CR9114''' || 8 || 15 || '''23''' || Mainly at the HA stem, including an HA2 helical region and adjacent loops/turns.
|-
| '''7K39, head-binding Fab''' || 14 || 1 || '''15''' || Mainly on HA1 in the head, distributed among accessible loops and neighboring structural elements.
|}

'''Figure placeholders:''' Upload the two PyMOL images to MediaWiki before enabling the lines below.


'''Why the previous 4FQI counts differed:''' The earlier PyMOL screenshot highlighted 18 residues (4 in A and 14 in B). A complete calculation with both antibody protein chains identifies another five within 5 Å: '''A:38, A:40, A:41, A:42 and B:52''', producing 23. An earlier estimate of 25 mistakenly allowed water atoms from the antibody-chain records to contribute contacts, adding A:18 and B:53. The protein-only, solvent-excluded result is '''23'''.

To reproduce the selection in PyMOL, for 4FQI:
<pre>
fetch 4fqi
remove solvent
select Ag, (chain A+B)
select Fab, (chain H+L)
select epitope, byres (Fab around 5)
color red, epitope
</pre>
For 7K39, replace the Fab selection with chain G+H

=== Q3. Which residues form the epitopes? Are they linear? Do they have a linear core? ===

'''4FQI (23 contacts):'''
* HA1, '''A''': 38, 40, 41, 42, 291, 292, 293, 318.
* HA2, '''B''': 18, 19, 20, 21, 36, 37, 38, 41, 42, 45, 46, 48, 49, 52, 56.

'''7K39 (15 contacts):'''
* HA1, '''A''': 48, 49, 50, 90, 91, 92, 93, 269, 270, 271, 272, 273, 284, 285.
* HA2, '''B''': 69.

'''Interpretation:''' Both are predominantly '''conformational (discontinuous) epitopes''', because the contacting residues occur in sequence-distant regions that come together in three dimensions. They contain short contiguous stretches or potential ''linear cores'', for example 4FQI B:18–21 and B:36–38, and 7K39 A:90–93 and A:269–273. These patches do not make the overall epitopes linear or show that the corresponding isolated peptides would bind the antibody.

The complexes do '''not''' target the same region: CR9114 binds the '''stem''' (involving HA1 and HA2), while the 7K39 antibody binds primarily the '''head''' (mostly HA1). HA1=head and HA2=stem are useful teaching approximations, but HA1 also contributes to the structural stem.

== Part B. Sequence-based prediction: BepiPred-2.0 and BepiPred-3.0 ==

'''Data limitation:''' The provided BepiPred-2.0 and BepiPred-3.0 CSV files contain predictions for the '''7K39 H3N2 HA sequences'''. They were not run on the native '''4FQI H5N1''' HA sequences. The two structures contain homologous but non-identical antigens. The quantitative answers in Q4–Q5 therefore directly test the '''7K39 antibody footprint only'''. To evaluate 4FQI independently, run both predictors on the 4FQI FASTA sequences.

The supplied prediction scores were thresholded at '''0.50''' (BepiPred-2.0) and '''0.1512''' (BepiPred-3.0 default, unsmoothed).

=== Q4. Do BepiPred-2.0 predictions overlap the actual epitopes? ===

'''Yes for 7K39, but the result is not very selective.''' All '''15/15''' experimentally observed 7K39 residues are predicted as epitope, giving '''100% recall'''. However, '''407/490''' evaluated residues ('''83.1%''') are predicted as positive, of which 392 are outside the particular observed antibody footprint. Precision is only '''15/407 = 3.69%'''.

A BepiPred-2.0 score of 0.5 therefore recovers the experimental footprint by labeling nearly the whole antigen as a potential epitope. This is not good evidence that it can pinpoint where the specific 7K39 antibody binds. Its score-ranked '''average precision (AP) is 0.0306''', approximately the random-ranking baseline '''15/490 = 0.0306'''.

=== Q5. Does the BepiPred-3.0 prediction overlap now? ===

'''Only slightly for 7K39 at the default threshold.''' Using the ''unsmoothed'' BepiPred-3.0 output, '''2/15''' experimental contact residues are predicted ('''13.3% recall'''): '''A:92''' and '''A:285'''. Their corresponding BepiPred sequence positions are 108 and 301. There are '''128/490''' predicted positive residues (26.1% coverage); precision is '''2/128 = 1.56%''', and '''AP = 0.0297'''.

The optionally ''smoothed'' BepiPred-3.0 score detects '''0/15''' experimental contacts at the same numerical threshold and labels 88/490 residues positive. Smoothing may emphasize contiguous patches but is not guaranteed to improve overlap with a discontinuous antibody footprint.

'''Conclusion for Q4–Q5:''' BepiPred-2.0 has higher recall at its default threshold but very poor localization, while BepiPred-3.0 makes fewer predictions but misses most of this particular antibody footprint. Neither is demonstrably good at ranking the 7K39 contact residues.

=== Q6. What is different in BepiPred-2.0 compared with BepiPred-3.0? ===

{| class="wikitable"
! Feature !! BepiPred-2.0 !! BepiPred-3.0
|-
| Core method || Random Forest using sequence-derived features, trained with structural epitope annotations || Neural networks using protein language-model (ESM-2) representations, trained with updated structural epitope annotations
|-
| Input || Protein sequence || Protein sequence
|-
| Output focus || Mainly continuous/linear epitope prediction with smoothing || Residue-level prediction for potential linear and discontinuous epitope residues, with optional smoothing for linear epitope analysis
|-
| Default threshold used here || 0.50 || 0.1512
|}

BepiPred-3.0 captures richer sequence context using learned embeddings, but neither tool receives the identity or structure of the specific antibody. Predictions are thus potential general B-cell epitope residues, not antibody-specific docking poses. The BepiPred-3.0 structural training annotation uses an antibody-contact cutoff of 4 Å, whereas this exercise uses 5 Å; this difference can also affect observed overlap.

See [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0] and [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0].

== Part C. Structure-based prediction: DiscoTope-2.0 and DiscoTope-3.0 ==

'''Data:''' The uploaded ''Discotope2pred.txt'' and ''file_A_discotope3.csv'' / ''file_B_discotope3.csv'' correspond to the '''7K39 antigen A/B'''. At the supplied settings, DiscoTope-2.0 calls positive residues with combined score greater than approximately '''−3.7'''; DiscoTope-3.0 uses its '''calibrated score ≥ 0.90''' (moderate confidence). The numerical scales are different and cannot be compared directly.

'''Input caveat:''' The DiscoTope-2.0 output includes other HA chains and Fab chains, indicating that the uploaded structure was not strictly antigen-only. DiscoTope-3.0 outputs were reported separately for HA chains A and B. The comparison is therefore useful for the supplied results, but not a fully controlled benchmark of the methods.

=== Q7. Does DiscoTope-2.0 overlap with the actual epitope? ===

'''No, not at the selected threshold for 7K39.''' DiscoTope-2.0 classifies '''58/490''' antigen residues as epitopes (34 in HA1 chain A; 24 in HA2 chain B). '''None of the 15''' experimentally observed 7K39 contact residues is predicted positive, giving '''recall = 0%''' and '''precision = 0%'''. The ranking-based '''AP = 0.0274''', slightly below the 0.0306 random-ranking baseline.

These findings are specific to the supplied output and settings; the method should ideally be rerun on the identical antigen-only structure supplied to the newer version.

=== Q8. Compare the results obtained using DiscoTope-2.0 and DiscoTope-3.0 ===

{| class="wikitable"
! Measurement for 7K39 HA A/B !! DiscoTope-2.0 !! DiscoTope-3.0
|-
| Model concept || Spatial epitope propensities plus residue exposure and neighborhood information || Learned structural representations, including inverse-folding features and exposure information
|-
| Threshold used || Combined score > −3.7 || Calibrated score ≥ 0.90
|-
| Residues predicted positive || 58/490 || 77/490
|-
| Experimental contact residues recovered || 0/15 || '''4/15'''
|-
| Recall || 0% || '''26.7%'''
|-
| Precision || 0% || '''5.19%'''
|-
| Average precision (AP) || 0.0274 || '''0.0466'''
|}

DiscoTope-3.0 correctly identifies experimental contacts '''A:91, A:93, A:271 and B:69'''. At the supplied cutoffs, the two versions share only '''12 predicted-positive HA residues'''. DiscoTope-3.0 therefore agrees better with the known 7K39 footprint in the supplied results, although it still misses 11 of the 15 contacts and predicts 73 residues outside this one interface.

See [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0] and [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0].

=== Q9. Which prediction best matches the experimentally observed epitope? ===

To answer this, compare both '''recall''' and '''precision''', rather than declaring any overlapping prediction correct.

<pre>
Precision = TP / (TP + FP)
Recall = TP / (TP + FN)
</pre>

Here TP denotes predicted residues in the measured 5 Å antibody footprint; FP denotes predicted residues outside that ''specific'' footprint; and FN denotes measured contact residues missed by the method. A threshold-independent score-ranking summary is '''average precision (AP)''', calculated from a precision–recall curve. Its random-ranking baseline here is '''15/490 = 3.06%'''.

{| class="wikitable sortable"
! Method (7K39) !! Correct contacts !! All predicted positives !! Recall !! Precision !! AP !! ROC-AUC
|-
| BepiPred-2.0 || 15/15 || 407/490 || '''100%''' || 3.69% || 0.0306 || 0.492
|-
| BepiPred-3.0, unsmoothed || 2/15 || 128/490 || 13.3% || 1.56% || 0.0297 || 0.492
|-
| DiscoTope-2.0 || 0/15 || 58/490 || 0% || 0% || 0.0274 || 0.413
|-
| '''DiscoTope-3.0''' || '''4/15''' || '''77/490''' || 26.7% || '''5.19%''' || '''0.0466''' || '''0.626'''
|}

'''Best result for this experimental interface: DiscoTope-3.0''', if “best” means highest precision and AP in the supplied data. '''BepiPred-2.0 has the highest recall''', but almost no ability to discriminate the actual interface from the rest of HA at its default threshold. All AP values remain low in absolute terms. The table is not a general benchmark because only one measured antibody footprint has directly matched predictor outputs and the DiscoTope inputs were not identical.



== Part D. Predicting where a known antibody binds (AlphaFold) ==

=== Q10. Are there tools that predict where an isolated antibody binds to an antigen? What do the provided AlphaFold models show? ===

'''Yes.''' Given antigen and antibody sequences, complex-modeling methods such as '''AlphaFold 3''' can propose a three-dimensional antibody–antigen arrangement. Related approaches include '''Boltz''' and docking methods such as '''HADDOCK'''. Unlike BepiPred or DiscoTope, these methods use information about the ''specific antibody'' and aim to model its candidate binding pose. A predicted pose is a hypothesis, not proof of binding.

The provided ZIP archives contain five AlphaFold 3 models each for a hemagglutinin trimer plus antibody chains. They illustrate both successful and problematic complex prediction:

{| class="wikitable"
! Feature !! 4FQI AlphaFold 3 archive !! 7K39 AlphaFold 3 archive
|-
| Experimentally observed epitope || Stem || Head
|-
| Predicted binding location || '''Stem''', consistent across the five models || '''Near the stem''', not the experimental head site; the HA protomer contacted changes in some models
|-
| Example model 0 predicted 5 Å contact residues || 24 (8 in HA1 A, 16 in HA2 B) || 31 (12 in HA1 C, 19 in HA2 D)
|-
| Global complex ipTM, model 0 || 0.85 || 0.62
|-
| Antibody–HA chain-pair ipTM, across supplied models || Approximately '''0.49–0.70''' || Only '''0.08–0.13'''
|-
| Interpretation || Plausible binding pose consistent with the known stem location || Poorly supported antibody–antigen arrangement, despite a moderately high global ipTM value
|}

'''Crucial caution:''' Global ipTM reflects multiple interfaces in a trimeric complex, including HA–HA interfaces. For determining whether the antibody is positioned correctly, examine '''antibody–antigen pairwise ipTM''', interchain predicted aligned error (PAE), and the actual molecular contact geometry. In 7K39, the model-0 minimum antibody–HA pairwise PAE values are about 20–25 Å, indicating considerable uncertainty in the relative placement. This explains why an apparently folded antibody and HA trimer can nevertheless be docked incorrectly.

'''Conclusion:''' The AlphaFold 3 models offer a binding-site hypothesis, and the 4FQI prediction agrees qualitatively with experiment. The 7K39 prediction fails to place the antibody at its experimental head site and has very low antibody–antigen interface confidence. These models should be evaluated by comparison with experimental structures and, where appropriate, binding assays, mutagenesis or other validation.

For confidence-score interpretation see [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 output documentation].

== Overall answer to the exercise ==

The known 4FQI stem and 7K39 head epitopes are spatially discontinuous and are associated with different antibody-binding regions of HA. Sequence-based BepiPred tools can recover candidate epitope residues but, for the supplied 7K39 predictions, do not discriminate the measured antibody footprint well. Of the four submitted predictions, DiscoTope-3.0 gives the strongest agreement with that particular 7K39 footprint (AP 0.0466 versus a random baseline of 0.0306), though the result is still modest. When the identity of an antibody is known, complex prediction can offer a more specific binding pose, but the 7K39 AlphaFold example demonstrates the risk of incorrect docking and the importance of interface confidence metrics. Predictions should be treated as testable hypotheses rather than definitive experimental epitopes.

== Limitations and reproducibility ==

* All experimental epitope counts here use a strict 5 Å '''protein-heavy-atom''' cutoff without solvent; 4FQI = '''23''' and 7K39 = '''15'''.
* BepiPred and DiscoTope numerical performance is directly measurable from the uploaded outputs for '''7K39 only'''. Native-H5 predictions for 4FQI were not supplied, so a proper 4FQI precision–recall evaluation cannot be completed using these outputs.
* The BepiPred residue positions must be translated to PDB numbering (7K39 A: minus 16; 7K39 B: unchanged).
* The DiscoTope-2.0 input includes more chains than the DiscoTope-3.0 antigen-only result, so their numerical comparison is potentially confounded.
* AP is a precision–recall ranking summary; it is not the same quantity as ROC-AUC. The positive class is only 15/490 residues, so conventional classification accuracy is not a useful primary metric.
* Results against only one experimental antibody interface cannot establish that other exposed residues are never epitopes for other antibodies.

== Sources and software ==

* [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction DTU 22145 exercise]
* [https://www.rcsb.org/structure/4FQI PDB 4FQI — H5N1 HA–CR9114]
* [https://www.rcsb.org/structure/7K39 PDB 7K39 — H3N2 HA–head-binding antibody]
* [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0]
* [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0]
* [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 confidence and output documentation]

Bcell Epitope prediction ans

2026-10-28T10:51:34Z

Carol: /* Q2. Show the epitopes, count residues, and describe secondary structure */

= B-cell epitope prediction =

== Analysis Tips and conventions ==

* '''Experimental epitope definition:''' An HA amino-acid residue belongs to the antibody-binding footprint if any of its non-hydrogen protein atoms is strictly less than 5.0 Å from a non-hydrogen protein atom of the selected antibody. Exclude waters and other non-protein molecules. Count a residue only once, even if several of its atoms contact the antibody.
* '''One interface at a time:''' For 7K39, use antigen chains A/B and one Fab (G/H); do not combine all three equivalent antibody copies.
* '''Residue numbering:''' The chain identifier and PDB residue number are part of the residue identity. For the supplied 7K39 BepiPred outputs, chain A sequence position = PDB residue number + 16; chain B sequence position = PDB residue number. This was confirmed by comparing the amino-acid sequences.
* '''Prediction evaluation:''' Predictions were compared on the 490 resolved amino acids in 7K39 HA chains A/B (318 in A and 172 in B). The default scores/flags were used unless stated otherwise.
* '''Important distinction:''' A residue outside one antibody's footprint is not necessarily a biologically impossible epitope; a different antibody could bind there. Thus “false positive” below means “outside the experimentally observed interface being tested.”

== Part A. Get your bearings in PyMOL (experimental epitopes) ==

=== Q1. List chain IDs for antigen versus antibody in each complex ===

{| class="wikitable"
! Experimental structure !! Antigen HA1 !! Antigen HA2 !! Antibody heavy chain !! Antibody light chain
|-
| '''4FQI''' (H5N1, CR9114) || A || B || H || L
|-
| '''7K39''' (H3N2, head-binding Fab) || A, C, E || B, D, F || H, J, L || G, I, K
|}

For the 7K39 calculations below use '''A/B''' as the antigen and '''G/H''' as a single Fab (G is light, H is heavy). The other equivalent antigen/Fab combinations are C/D with I/J, and E/F with K/L.

'''Note:''' The AlphaFold 3 ZIP files use their own chain labels. Do not transfer the experimental PDB chain mapping to the predicted complexes without checking their sequences.

=== Q2. Show the epitopes, count residues, and describe secondary structure ===

Using the experimental CIF coordinates, excluding waters, and applying the 5 Å protein-heavy-atom cutoff:

{| class="wikitable"
! Complex !! HA1 (A) contact residues !! HA2 (B) contact residues !! Total !! Qualitative structural location
|-
| '''4FQI, stem-binding CR9114''' || 8 || 15 || '''23''' || Mainly at the HA stem, including an HA2 helical region and adjacent loops/turns.
|-
| '''7K39, head-binding Fab''' || 14 || 1 || '''15''' || Mainly on HA1 in the head, distributed among accessible loops and neighboring structural elements.
|}

'''Figure placeholders:''' Upload the two PyMOL images to MediaWiki before enabling the lines below.


'''Why the previous 4FQI counts differed:''' The earlier PyMOL screenshot highlighted 18 residues (4 in A and 14 in B). A complete calculation with both antibody protein chains identifies another five within 5 Å: '''A:38, A:40, A:41, A:42 and B:52''', producing 23. An earlier estimate of 25 mistakenly allowed water atoms from the antibody-chain records to contribute contacts, adding A:18 and B:53. The protein-only, solvent-excluded result is '''23'''.

To reproduce the selection in PyMOL, for 4FQI:
<pre>
fetch 4fqi
remove solvent
select Ag, (chain A+B)
select Fab, (chain H+L)
select epitope, byres (Fab around 5)
color red, epitope
</pre>
For 7K39, replace the Fab selection with <code>(chain G+H) and polymer.protein</code> and keep A/B as the chosen antigen protomer.

=== Q3. Which residues form the epitopes? Are they linear? Do they have a linear core? ===

'''4FQI (23 contacts):'''
* HA1, '''A''': 38, 40, 41, 42, 291, 292, 293, 318.
* HA2, '''B''': 18, 19, 20, 21, 36, 37, 38, 41, 42, 45, 46, 48, 49, 52, 56.

'''7K39 (15 contacts):'''
* HA1, '''A''': 48, 49, 50, 90, 91, 92, 93, 269, 270, 271, 272, 273, 284, 285.
* HA2, '''B''': 69.

'''Interpretation:''' Both are predominantly '''conformational (discontinuous) epitopes''', because the contacting residues occur in sequence-distant regions that come together in three dimensions. They contain short contiguous stretches or potential ''linear cores'', for example 4FQI B:18–21 and B:36–38, and 7K39 A:90–93 and A:269–273. These patches do not make the overall epitopes linear or show that the corresponding isolated peptides would bind the antibody.

The complexes do '''not''' target the same region: CR9114 binds the '''stem''' (involving HA1 and HA2), while the 7K39 antibody binds primarily the '''head''' (mostly HA1). HA1=head and HA2=stem are useful teaching approximations, but HA1 also contributes to the structural stem.

== Part B. Sequence-based prediction: BepiPred-2.0 and BepiPred-3.0 ==

'''Data limitation:''' The provided BepiPred-2.0 and BepiPred-3.0 CSV files contain predictions for the '''7K39 H3N2 HA sequences'''. They were not run on the native '''4FQI H5N1''' HA sequences. The two structures contain homologous but non-identical antigens. The quantitative answers in Q4–Q5 therefore directly test the '''7K39 antibody footprint only'''. To evaluate 4FQI independently, run both predictors on the 4FQI FASTA sequences.

The supplied prediction scores were thresholded at '''0.50''' (BepiPred-2.0) and '''0.1512''' (BepiPred-3.0 default, unsmoothed).

=== Q4. Do BepiPred-2.0 predictions overlap the actual epitopes? ===

'''Yes for 7K39, but the result is not very selective.''' All '''15/15''' experimentally observed 7K39 residues are predicted as epitope, giving '''100% recall'''. However, '''407/490''' evaluated residues ('''83.1%''') are predicted as positive, of which 392 are outside the particular observed antibody footprint. Precision is only '''15/407 = 3.69%'''.

A BepiPred-2.0 score of 0.5 therefore recovers the experimental footprint by labeling nearly the whole antigen as a potential epitope. This is not good evidence that it can pinpoint where the specific 7K39 antibody binds. Its score-ranked '''average precision (AP) is 0.0306''', approximately the random-ranking baseline '''15/490 = 0.0306'''.

=== Q5. Does the BepiPred-3.0 prediction overlap now? ===

'''Only slightly for 7K39 at the default threshold.''' Using the ''unsmoothed'' BepiPred-3.0 output, '''2/15''' experimental contact residues are predicted ('''13.3% recall'''): '''A:92''' and '''A:285'''. Their corresponding BepiPred sequence positions are 108 and 301. There are '''128/490''' predicted positive residues (26.1% coverage); precision is '''2/128 = 1.56%''', and '''AP = 0.0297'''.

The optionally ''smoothed'' BepiPred-3.0 score detects '''0/15''' experimental contacts at the same numerical threshold and labels 88/490 residues positive. Smoothing may emphasize contiguous patches but is not guaranteed to improve overlap with a discontinuous antibody footprint.

'''Conclusion for Q4–Q5:''' BepiPred-2.0 has higher recall at its default threshold but very poor localization, while BepiPred-3.0 makes fewer predictions but misses most of this particular antibody footprint. Neither is demonstrably good at ranking the 7K39 contact residues.

=== Q6. What is different in BepiPred-2.0 compared with BepiPred-3.0? ===

{| class="wikitable"
! Feature !! BepiPred-2.0 !! BepiPred-3.0
|-
| Core method || Random Forest using sequence-derived features, trained with structural epitope annotations || Neural networks using protein language-model (ESM-2) representations, trained with updated structural epitope annotations
|-
| Input || Protein sequence || Protein sequence
|-
| Output focus || Mainly continuous/linear epitope prediction with smoothing || Residue-level prediction for potential linear and discontinuous epitope residues, with optional smoothing for linear epitope analysis
|-
| Default threshold used here || 0.50 || 0.1512
|}

BepiPred-3.0 captures richer sequence context using learned embeddings, but neither tool receives the identity or structure of the specific antibody. Predictions are thus potential general B-cell epitope residues, not antibody-specific docking poses. The BepiPred-3.0 structural training annotation uses an antibody-contact cutoff of 4 Å, whereas this exercise uses 5 Å; this difference can also affect observed overlap.

See [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0] and [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0].

== Part C. Structure-based prediction: DiscoTope-2.0 and DiscoTope-3.0 ==

'''Data:''' The uploaded ''Discotope2pred.txt'' and ''file_A_discotope3.csv'' / ''file_B_discotope3.csv'' correspond to the '''7K39 antigen A/B'''. At the supplied settings, DiscoTope-2.0 calls positive residues with combined score greater than approximately '''−3.7'''; DiscoTope-3.0 uses its '''calibrated score ≥ 0.90''' (moderate confidence). The numerical scales are different and cannot be compared directly.

'''Input caveat:''' The DiscoTope-2.0 output includes other HA chains and Fab chains, indicating that the uploaded structure was not strictly antigen-only. DiscoTope-3.0 outputs were reported separately for HA chains A and B. The comparison is therefore useful for the supplied results, but not a fully controlled benchmark of the methods.

=== Q7. Does DiscoTope-2.0 overlap with the actual epitope? ===

'''No, not at the selected threshold for 7K39.''' DiscoTope-2.0 classifies '''58/490''' antigen residues as epitopes (34 in HA1 chain A; 24 in HA2 chain B). '''None of the 15''' experimentally observed 7K39 contact residues is predicted positive, giving '''recall = 0%''' and '''precision = 0%'''. The ranking-based '''AP = 0.0274''', slightly below the 0.0306 random-ranking baseline.

These findings are specific to the supplied output and settings; the method should ideally be rerun on the identical antigen-only structure supplied to the newer version.

=== Q8. Compare the results obtained using DiscoTope-2.0 and DiscoTope-3.0 ===

{| class="wikitable"
! Measurement for 7K39 HA A/B !! DiscoTope-2.0 !! DiscoTope-3.0
|-
| Model concept || Spatial epitope propensities plus residue exposure and neighborhood information || Learned structural representations, including inverse-folding features and exposure information
|-
| Threshold used || Combined score > −3.7 || Calibrated score ≥ 0.90
|-
| Residues predicted positive || 58/490 || 77/490
|-
| Experimental contact residues recovered || 0/15 || '''4/15'''
|-
| Recall || 0% || '''26.7%'''
|-
| Precision || 0% || '''5.19%'''
|-
| Average precision (AP) || 0.0274 || '''0.0466'''
|}

DiscoTope-3.0 correctly identifies experimental contacts '''A:91, A:93, A:271 and B:69'''. At the supplied cutoffs, the two versions share only '''12 predicted-positive HA residues'''. DiscoTope-3.0 therefore agrees better with the known 7K39 footprint in the supplied results, although it still misses 11 of the 15 contacts and predicts 73 residues outside this one interface.

See [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0] and [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0].

=== Q9. Which prediction best matches the experimentally observed epitope? ===

To answer this, compare both '''recall''' and '''precision''', rather than declaring any overlapping prediction correct.

<pre>
Precision = TP / (TP + FP)
Recall = TP / (TP + FN)
</pre>

Here TP denotes predicted residues in the measured 5 Å antibody footprint; FP denotes predicted residues outside that ''specific'' footprint; and FN denotes measured contact residues missed by the method. A threshold-independent score-ranking summary is '''average precision (AP)''', calculated from a precision–recall curve. Its random-ranking baseline here is '''15/490 = 3.06%'''.

{| class="wikitable sortable"
! Method (7K39) !! Correct contacts !! All predicted positives !! Recall !! Precision !! AP !! ROC-AUC
|-
| BepiPred-2.0 || 15/15 || 407/490 || '''100%''' || 3.69% || 0.0306 || 0.492
|-
| BepiPred-3.0, unsmoothed || 2/15 || 128/490 || 13.3% || 1.56% || 0.0297 || 0.492
|-
| DiscoTope-2.0 || 0/15 || 58/490 || 0% || 0% || 0.0274 || 0.413
|-
| '''DiscoTope-3.0''' || '''4/15''' || '''77/490''' || 26.7% || '''5.19%''' || '''0.0466''' || '''0.626'''
|}

'''Best result for this experimental interface: DiscoTope-3.0''', if “best” means highest precision and AP in the supplied data. '''BepiPred-2.0 has the highest recall''', but almost no ability to discriminate the actual interface from the rest of HA at its default threshold. All AP values remain low in absolute terms. The table is not a general benchmark because only one measured antibody footprint has directly matched predictor outputs and the DiscoTope inputs were not identical.



== Part D. Predicting where a known antibody binds (AlphaFold) ==

=== Q10. Are there tools that predict where an isolated antibody binds to an antigen? What do the provided AlphaFold models show? ===

'''Yes.''' Given antigen and antibody sequences, complex-modeling methods such as '''AlphaFold 3''' can propose a three-dimensional antibody–antigen arrangement. Related approaches include '''Boltz''' and docking methods such as '''HADDOCK'''. Unlike BepiPred or DiscoTope, these methods use information about the ''specific antibody'' and aim to model its candidate binding pose. A predicted pose is a hypothesis, not proof of binding.

The provided ZIP archives contain five AlphaFold 3 models each for a hemagglutinin trimer plus antibody chains. They illustrate both successful and problematic complex prediction:

{| class="wikitable"
! Feature !! 4FQI AlphaFold 3 archive !! 7K39 AlphaFold 3 archive
|-
| Experimentally observed epitope || Stem || Head
|-
| Predicted binding location || '''Stem''', consistent across the five models || '''Near the stem''', not the experimental head site; the HA protomer contacted changes in some models
|-
| Example model 0 predicted 5 Å contact residues || 24 (8 in HA1 A, 16 in HA2 B) || 31 (12 in HA1 C, 19 in HA2 D)
|-
| Global complex ipTM, model 0 || 0.85 || 0.62
|-
| Antibody–HA chain-pair ipTM, across supplied models || Approximately '''0.49–0.70''' || Only '''0.08–0.13'''
|-
| Interpretation || Plausible binding pose consistent with the known stem location || Poorly supported antibody–antigen arrangement, despite a moderately high global ipTM value
|}

'''Crucial caution:''' Global ipTM reflects multiple interfaces in a trimeric complex, including HA–HA interfaces. For determining whether the antibody is positioned correctly, examine '''antibody–antigen pairwise ipTM''', interchain predicted aligned error (PAE), and the actual molecular contact geometry. In 7K39, the model-0 minimum antibody–HA pairwise PAE values are about 20–25 Å, indicating considerable uncertainty in the relative placement. This explains why an apparently folded antibody and HA trimer can nevertheless be docked incorrectly.

'''Conclusion:''' The AlphaFold 3 models offer a binding-site hypothesis, and the 4FQI prediction agrees qualitatively with experiment. The 7K39 prediction fails to place the antibody at its experimental head site and has very low antibody–antigen interface confidence. These models should be evaluated by comparison with experimental structures and, where appropriate, binding assays, mutagenesis or other validation.

For confidence-score interpretation see [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 output documentation].

== Overall answer to the exercise ==

The known 4FQI stem and 7K39 head epitopes are spatially discontinuous and are associated with different antibody-binding regions of HA. Sequence-based BepiPred tools can recover candidate epitope residues but, for the supplied 7K39 predictions, do not discriminate the measured antibody footprint well. Of the four submitted predictions, DiscoTope-3.0 gives the strongest agreement with that particular 7K39 footprint (AP 0.0466 versus a random baseline of 0.0306), though the result is still modest. When the identity of an antibody is known, complex prediction can offer a more specific binding pose, but the 7K39 AlphaFold example demonstrates the risk of incorrect docking and the importance of interface confidence metrics. Predictions should be treated as testable hypotheses rather than definitive experimental epitopes.

== Limitations and reproducibility ==

* All experimental epitope counts here use a strict 5 Å '''protein-heavy-atom''' cutoff without solvent; 4FQI = '''23''' and 7K39 = '''15'''.
* BepiPred and DiscoTope numerical performance is directly measurable from the uploaded outputs for '''7K39 only'''. Native-H5 predictions for 4FQI were not supplied, so a proper 4FQI precision–recall evaluation cannot be completed using these outputs.
* The BepiPred residue positions must be translated to PDB numbering (7K39 A: minus 16; 7K39 B: unchanged).
* The DiscoTope-2.0 input includes more chains than the DiscoTope-3.0 antigen-only result, so their numerical comparison is potentially confounded.
* AP is a precision–recall ranking summary; it is not the same quantity as ROC-AUC. The positive class is only 15/490 residues, so conventional classification accuracy is not a useful primary metric.
* Results against only one experimental antibody interface cannot establish that other exposed residues are never epitopes for other antibodies.

== Sources and software ==

* [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction DTU 22145 exercise]
* [https://www.rcsb.org/structure/4FQI PDB 4FQI — H5N1 HA–CR9114]
* [https://www.rcsb.org/structure/7K39 PDB 7K39 — H3N2 HA–head-binding antibody]
* [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0]
* [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0]
* [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 confidence and output documentation]

Bcell Epitope prediction ans

2026-10-28T10:50:10Z

Carol: /* B-cell epitope prediction — Instructor answer key */

= B-cell epitope prediction =

== Analysis Tips and conventions ==

* '''Experimental epitope definition:''' An HA amino-acid residue belongs to the antibody-binding footprint if any of its non-hydrogen protein atoms is strictly less than 5.0 Å from a non-hydrogen protein atom of the selected antibody. Exclude waters and other non-protein molecules. Count a residue only once, even if several of its atoms contact the antibody.
* '''One interface at a time:''' For 7K39, use antigen chains A/B and one Fab (G/H); do not combine all three equivalent antibody copies.
* '''Residue numbering:''' The chain identifier and PDB residue number are part of the residue identity. For the supplied 7K39 BepiPred outputs, chain A sequence position = PDB residue number + 16; chain B sequence position = PDB residue number. This was confirmed by comparing the amino-acid sequences.
* '''Prediction evaluation:''' Predictions were compared on the 490 resolved amino acids in 7K39 HA chains A/B (318 in A and 172 in B). The default scores/flags were used unless stated otherwise.
* '''Important distinction:''' A residue outside one antibody's footprint is not necessarily a biologically impossible epitope; a different antibody could bind there. Thus “false positive” below means “outside the experimentally observed interface being tested.”

== Part A. Get your bearings in PyMOL (experimental epitopes) ==

=== Q1. List chain IDs for antigen versus antibody in each complex ===

{| class="wikitable"
! Experimental structure !! Antigen HA1 !! Antigen HA2 !! Antibody heavy chain !! Antibody light chain
|-
| '''4FQI''' (H5N1, CR9114) || A || B || H || L
|-
| '''7K39''' (H3N2, head-binding Fab) || A, C, E || B, D, F || H, J, L || G, I, K
|}

For the 7K39 calculations below use '''A/B''' as the antigen and '''G/H''' as a single Fab (G is light, H is heavy). The other equivalent antigen/Fab combinations are C/D with I/J, and E/F with K/L.

'''Note:''' The AlphaFold 3 ZIP files use their own chain labels. Do not transfer the experimental PDB chain mapping to the predicted complexes without checking their sequences.

=== Q2. Show the epitopes, count residues, and describe secondary structure ===

Using the experimental CIF coordinates, excluding waters, and applying the 5 Å protein-heavy-atom cutoff:

{| class="wikitable"
! Complex !! HA1 (A) contact residues !! HA2 (B) contact residues !! Total !! Qualitative structural location
|-
| '''4FQI, stem-binding CR9114''' || 8 || 15 || '''23''' || Mainly at the HA stem, including an HA2 helical region and adjacent loops/turns.
|-
| '''7K39, head-binding Fab''' || 14 || 1 || '''15''' || Mainly on HA1 in the head, distributed among accessible loops and neighboring structural elements.
|}

'''Figure placeholders:''' Upload the two PyMOL images to MediaWiki before enabling the lines below.


'''Why the previous 4FQI counts differed:''' The earlier PyMOL screenshot highlighted 18 residues (4 in A and 14 in B). A complete calculation with both antibody protein chains identifies another five within 5 Å: '''A:38, A:40, A:41, A:42 and B:52''', producing 23. An earlier estimate of 25 mistakenly allowed water atoms from the antibody-chain records to contribute contacts, adding A:18 and B:53. The protein-only, solvent-excluded result is '''23'''.

To reproduce the selection in PyMOL, for 4FQI:
<pre>
fetch 4fqi
remove solvent
select ag, (chain A+B) and polymer.protein
select ab, (chain H+L) and polymer.protein
select epitope, byres (ag within 5 of ab)
color red, epitope
</pre>
For 7K39, replace the Fab selection with <code>(chain G+H) and polymer.protein</code> and keep A/B as the chosen antigen protomer.

=== Q3. Which residues form the epitopes? Are they linear? Do they have a linear core? ===

'''4FQI (23 contacts):'''
* HA1, '''A''': 38, 40, 41, 42, 291, 292, 293, 318.
* HA2, '''B''': 18, 19, 20, 21, 36, 37, 38, 41, 42, 45, 46, 48, 49, 52, 56.

'''7K39 (15 contacts):'''
* HA1, '''A''': 48, 49, 50, 90, 91, 92, 93, 269, 270, 271, 272, 273, 284, 285.
* HA2, '''B''': 69.

'''Interpretation:''' Both are predominantly '''conformational (discontinuous) epitopes''', because the contacting residues occur in sequence-distant regions that come together in three dimensions. They contain short contiguous stretches or potential ''linear cores'', for example 4FQI B:18–21 and B:36–38, and 7K39 A:90–93 and A:269–273. These patches do not make the overall epitopes linear or show that the corresponding isolated peptides would bind the antibody.

The complexes do '''not''' target the same region: CR9114 binds the '''stem''' (involving HA1 and HA2), while the 7K39 antibody binds primarily the '''head''' (mostly HA1). HA1=head and HA2=stem are useful teaching approximations, but HA1 also contributes to the structural stem.

== Part B. Sequence-based prediction: BepiPred-2.0 and BepiPred-3.0 ==

'''Data limitation:''' The provided BepiPred-2.0 and BepiPred-3.0 CSV files contain predictions for the '''7K39 H3N2 HA sequences'''. They were not run on the native '''4FQI H5N1''' HA sequences. The two structures contain homologous but non-identical antigens. The quantitative answers in Q4–Q5 therefore directly test the '''7K39 antibody footprint only'''. To evaluate 4FQI independently, run both predictors on the 4FQI FASTA sequences.

The supplied prediction scores were thresholded at '''0.50''' (BepiPred-2.0) and '''0.1512''' (BepiPred-3.0 default, unsmoothed).

=== Q4. Do BepiPred-2.0 predictions overlap the actual epitopes? ===

'''Yes for 7K39, but the result is not very selective.''' All '''15/15''' experimentally observed 7K39 residues are predicted as epitope, giving '''100% recall'''. However, '''407/490''' evaluated residues ('''83.1%''') are predicted as positive, of which 392 are outside the particular observed antibody footprint. Precision is only '''15/407 = 3.69%'''.

A BepiPred-2.0 score of 0.5 therefore recovers the experimental footprint by labeling nearly the whole antigen as a potential epitope. This is not good evidence that it can pinpoint where the specific 7K39 antibody binds. Its score-ranked '''average precision (AP) is 0.0306''', approximately the random-ranking baseline '''15/490 = 0.0306'''.

=== Q5. Does the BepiPred-3.0 prediction overlap now? ===

'''Only slightly for 7K39 at the default threshold.''' Using the ''unsmoothed'' BepiPred-3.0 output, '''2/15''' experimental contact residues are predicted ('''13.3% recall'''): '''A:92''' and '''A:285'''. Their corresponding BepiPred sequence positions are 108 and 301. There are '''128/490''' predicted positive residues (26.1% coverage); precision is '''2/128 = 1.56%''', and '''AP = 0.0297'''.

The optionally ''smoothed'' BepiPred-3.0 score detects '''0/15''' experimental contacts at the same numerical threshold and labels 88/490 residues positive. Smoothing may emphasize contiguous patches but is not guaranteed to improve overlap with a discontinuous antibody footprint.

'''Conclusion for Q4–Q5:''' BepiPred-2.0 has higher recall at its default threshold but very poor localization, while BepiPred-3.0 makes fewer predictions but misses most of this particular antibody footprint. Neither is demonstrably good at ranking the 7K39 contact residues.

=== Q6. What is different in BepiPred-2.0 compared with BepiPred-3.0? ===

{| class="wikitable"
! Feature !! BepiPred-2.0 !! BepiPred-3.0
|-
| Core method || Random Forest using sequence-derived features, trained with structural epitope annotations || Neural networks using protein language-model (ESM-2) representations, trained with updated structural epitope annotations
|-
| Input || Protein sequence || Protein sequence
|-
| Output focus || Mainly continuous/linear epitope prediction with smoothing || Residue-level prediction for potential linear and discontinuous epitope residues, with optional smoothing for linear epitope analysis
|-
| Default threshold used here || 0.50 || 0.1512
|}

BepiPred-3.0 captures richer sequence context using learned embeddings, but neither tool receives the identity or structure of the specific antibody. Predictions are thus potential general B-cell epitope residues, not antibody-specific docking poses. The BepiPred-3.0 structural training annotation uses an antibody-contact cutoff of 4 Å, whereas this exercise uses 5 Å; this difference can also affect observed overlap.

See [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0] and [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0].

== Part C. Structure-based prediction: DiscoTope-2.0 and DiscoTope-3.0 ==

'''Data:''' The uploaded ''Discotope2pred.txt'' and ''file_A_discotope3.csv'' / ''file_B_discotope3.csv'' correspond to the '''7K39 antigen A/B'''. At the supplied settings, DiscoTope-2.0 calls positive residues with combined score greater than approximately '''−3.7'''; DiscoTope-3.0 uses its '''calibrated score ≥ 0.90''' (moderate confidence). The numerical scales are different and cannot be compared directly.

'''Input caveat:''' The DiscoTope-2.0 output includes other HA chains and Fab chains, indicating that the uploaded structure was not strictly antigen-only. DiscoTope-3.0 outputs were reported separately for HA chains A and B. The comparison is therefore useful for the supplied results, but not a fully controlled benchmark of the methods.

=== Q7. Does DiscoTope-2.0 overlap with the actual epitope? ===

'''No, not at the selected threshold for 7K39.''' DiscoTope-2.0 classifies '''58/490''' antigen residues as epitopes (34 in HA1 chain A; 24 in HA2 chain B). '''None of the 15''' experimentally observed 7K39 contact residues is predicted positive, giving '''recall = 0%''' and '''precision = 0%'''. The ranking-based '''AP = 0.0274''', slightly below the 0.0306 random-ranking baseline.

These findings are specific to the supplied output and settings; the method should ideally be rerun on the identical antigen-only structure supplied to the newer version.

=== Q8. Compare the results obtained using DiscoTope-2.0 and DiscoTope-3.0 ===

{| class="wikitable"
! Measurement for 7K39 HA A/B !! DiscoTope-2.0 !! DiscoTope-3.0
|-
| Model concept || Spatial epitope propensities plus residue exposure and neighborhood information || Learned structural representations, including inverse-folding features and exposure information
|-
| Threshold used || Combined score > −3.7 || Calibrated score ≥ 0.90
|-
| Residues predicted positive || 58/490 || 77/490
|-
| Experimental contact residues recovered || 0/15 || '''4/15'''
|-
| Recall || 0% || '''26.7%'''
|-
| Precision || 0% || '''5.19%'''
|-
| Average precision (AP) || 0.0274 || '''0.0466'''
|}

DiscoTope-3.0 correctly identifies experimental contacts '''A:91, A:93, A:271 and B:69'''. At the supplied cutoffs, the two versions share only '''12 predicted-positive HA residues'''. DiscoTope-3.0 therefore agrees better with the known 7K39 footprint in the supplied results, although it still misses 11 of the 15 contacts and predicts 73 residues outside this one interface.

See [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0] and [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0].

=== Q9. Which prediction best matches the experimentally observed epitope? ===

To answer this, compare both '''recall''' and '''precision''', rather than declaring any overlapping prediction correct.

<pre>
Precision = TP / (TP + FP)
Recall = TP / (TP + FN)
</pre>

Here TP denotes predicted residues in the measured 5 Å antibody footprint; FP denotes predicted residues outside that ''specific'' footprint; and FN denotes measured contact residues missed by the method. A threshold-independent score-ranking summary is '''average precision (AP)''', calculated from a precision–recall curve. Its random-ranking baseline here is '''15/490 = 3.06%'''.

{| class="wikitable sortable"
! Method (7K39) !! Correct contacts !! All predicted positives !! Recall !! Precision !! AP !! ROC-AUC
|-
| BepiPred-2.0 || 15/15 || 407/490 || '''100%''' || 3.69% || 0.0306 || 0.492
|-
| BepiPred-3.0, unsmoothed || 2/15 || 128/490 || 13.3% || 1.56% || 0.0297 || 0.492
|-
| DiscoTope-2.0 || 0/15 || 58/490 || 0% || 0% || 0.0274 || 0.413
|-
| '''DiscoTope-3.0''' || '''4/15''' || '''77/490''' || 26.7% || '''5.19%''' || '''0.0466''' || '''0.626'''
|}

'''Best result for this experimental interface: DiscoTope-3.0''', if “best” means highest precision and AP in the supplied data. '''BepiPred-2.0 has the highest recall''', but almost no ability to discriminate the actual interface from the rest of HA at its default threshold. All AP values remain low in absolute terms. The table is not a general benchmark because only one measured antibody footprint has directly matched predictor outputs and the DiscoTope inputs were not identical.



== Part D. Predicting where a known antibody binds (AlphaFold) ==

=== Q10. Are there tools that predict where an isolated antibody binds to an antigen? What do the provided AlphaFold models show? ===

'''Yes.''' Given antigen and antibody sequences, complex-modeling methods such as '''AlphaFold 3''' can propose a three-dimensional antibody–antigen arrangement. Related approaches include '''Boltz''' and docking methods such as '''HADDOCK'''. Unlike BepiPred or DiscoTope, these methods use information about the ''specific antibody'' and aim to model its candidate binding pose. A predicted pose is a hypothesis, not proof of binding.

The provided ZIP archives contain five AlphaFold 3 models each for a hemagglutinin trimer plus antibody chains. They illustrate both successful and problematic complex prediction:

{| class="wikitable"
! Feature !! 4FQI AlphaFold 3 archive !! 7K39 AlphaFold 3 archive
|-
| Experimentally observed epitope || Stem || Head
|-
| Predicted binding location || '''Stem''', consistent across the five models || '''Near the stem''', not the experimental head site; the HA protomer contacted changes in some models
|-
| Example model 0 predicted 5 Å contact residues || 24 (8 in HA1 A, 16 in HA2 B) || 31 (12 in HA1 C, 19 in HA2 D)
|-
| Global complex ipTM, model 0 || 0.85 || 0.62
|-
| Antibody–HA chain-pair ipTM, across supplied models || Approximately '''0.49–0.70''' || Only '''0.08–0.13'''
|-
| Interpretation || Plausible binding pose consistent with the known stem location || Poorly supported antibody–antigen arrangement, despite a moderately high global ipTM value
|}

'''Crucial caution:''' Global ipTM reflects multiple interfaces in a trimeric complex, including HA–HA interfaces. For determining whether the antibody is positioned correctly, examine '''antibody–antigen pairwise ipTM''', interchain predicted aligned error (PAE), and the actual molecular contact geometry. In 7K39, the model-0 minimum antibody–HA pairwise PAE values are about 20–25 Å, indicating considerable uncertainty in the relative placement. This explains why an apparently folded antibody and HA trimer can nevertheless be docked incorrectly.

'''Conclusion:''' The AlphaFold 3 models offer a binding-site hypothesis, and the 4FQI prediction agrees qualitatively with experiment. The 7K39 prediction fails to place the antibody at its experimental head site and has very low antibody–antigen interface confidence. These models should be evaluated by comparison with experimental structures and, where appropriate, binding assays, mutagenesis or other validation.

For confidence-score interpretation see [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 output documentation].

== Overall answer to the exercise ==

The known 4FQI stem and 7K39 head epitopes are spatially discontinuous and are associated with different antibody-binding regions of HA. Sequence-based BepiPred tools can recover candidate epitope residues but, for the supplied 7K39 predictions, do not discriminate the measured antibody footprint well. Of the four submitted predictions, DiscoTope-3.0 gives the strongest agreement with that particular 7K39 footprint (AP 0.0466 versus a random baseline of 0.0306), though the result is still modest. When the identity of an antibody is known, complex prediction can offer a more specific binding pose, but the 7K39 AlphaFold example demonstrates the risk of incorrect docking and the importance of interface confidence metrics. Predictions should be treated as testable hypotheses rather than definitive experimental epitopes.

== Limitations and reproducibility ==

* All experimental epitope counts here use a strict 5 Å '''protein-heavy-atom''' cutoff without solvent; 4FQI = '''23''' and 7K39 = '''15'''.
* BepiPred and DiscoTope numerical performance is directly measurable from the uploaded outputs for '''7K39 only'''. Native-H5 predictions for 4FQI were not supplied, so a proper 4FQI precision–recall evaluation cannot be completed using these outputs.
* The BepiPred residue positions must be translated to PDB numbering (7K39 A: minus 16; 7K39 B: unchanged).
* The DiscoTope-2.0 input includes more chains than the DiscoTope-3.0 antigen-only result, so their numerical comparison is potentially confounded.
* AP is a precision–recall ranking summary; it is not the same quantity as ROC-AUC. The positive class is only 15/490 residues, so conventional classification accuracy is not a useful primary metric.
* Results against only one experimental antibody interface cannot establish that other exposed residues are never epitopes for other antibodies.

== Sources and software ==

* [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction DTU 22145 exercise]
* [https://www.rcsb.org/structure/4FQI PDB 4FQI — H5N1 HA–CR9114]
* [https://www.rcsb.org/structure/7K39 PDB 7K39 — H3N2 HA–head-binding antibody]
* [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0]
* [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0]
* [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 confidence and output documentation]

Bcell Epitope prediction ans

2026-10-28T10:48:30Z

Carol: Created page with "<!-- INSTRUCTOR ANSWER KEY — DTU 22145 Matches the CURRENT exercise page checked 28 October 2026 (Q1–Q10; Parts A–D). Derived from uploaded experimental 4FQI.cif and 7K39.cif; supplied BepiPred-2, BepiPred-3, DiscoTope-2 and DiscoTope-3 files for 7K39; and AlphaFold 3 prediction archives for 4FQI and 7K39. This file is intended for an instructor page; it reveals the answers. The prediction performance below is for 7K39 H3N2 only, unless stated. Do not interpret res..."

= B-cell epitope prediction — Instructor answer key =

''Worked solution to the [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction DTU 22145 exercise]. The experimental structures are [https://www.rcsb.org/structure/4FQI 4FQI] (H5N1 influenza HA with stem-binding antibody CR9114) and [https://www.rcsb.org/structure/7K39 7K39] (H3N2 HA with a head-binding antibody). Numbers refer to the supplied structural and prediction files.''

== Analysis conventions ==

* '''Experimental epitope definition:''' An HA amino-acid residue belongs to the antibody-binding footprint if any of its non-hydrogen protein atoms is strictly less than 5.0 Å from a non-hydrogen protein atom of the selected antibody. Exclude waters and other non-protein molecules. Count a residue only once, even if several of its atoms contact the antibody.
* '''One interface at a time:''' For 7K39, use antigen chains A/B and one Fab (G/H); do not combine all three equivalent antibody copies.
* '''Residue numbering:''' The chain identifier and PDB residue number are part of the residue identity. For the supplied 7K39 BepiPred outputs, chain A sequence position = PDB residue number + 16; chain B sequence position = PDB residue number. This was confirmed by comparing the amino-acid sequences.
* '''Prediction evaluation:''' Predictions were compared on the 490 resolved amino acids in 7K39 HA chains A/B (318 in A and 172 in B). The default scores/flags were used unless stated otherwise.
* '''Important distinction:''' A residue outside one antibody's footprint is not necessarily a biologically impossible epitope; a different antibody could bind there. Thus “false positive” below means “outside the experimentally observed interface being tested.”

== Part A. Get your bearings in PyMOL (experimental epitopes) ==

=== Q1. List chain IDs for antigen versus antibody in each complex ===

{| class="wikitable"
! Experimental structure !! Antigen HA1 !! Antigen HA2 !! Antibody heavy chain !! Antibody light chain
|-
| '''4FQI''' (H5N1, CR9114) || A || B || H || L
|-
| '''7K39''' (H3N2, head-binding Fab) || A, C, E || B, D, F || H, J, L || G, I, K
|}

For the 7K39 calculations below use '''A/B''' as the antigen and '''G/H''' as a single Fab (G is light, H is heavy). The other equivalent antigen/Fab combinations are C/D with I/J, and E/F with K/L.

'''Note:''' The AlphaFold 3 ZIP files use their own chain labels. Do not transfer the experimental PDB chain mapping to the predicted complexes without checking their sequences.

=== Q2. Show the epitopes, count residues, and describe secondary structure ===

Using the experimental CIF coordinates, excluding waters, and applying the 5 Å protein-heavy-atom cutoff:

{| class="wikitable"
! Complex !! HA1 (A) contact residues !! HA2 (B) contact residues !! Total !! Qualitative structural location
|-
| '''4FQI, stem-binding CR9114''' || 8 || 15 || '''23''' || Mainly at the HA stem, including an HA2 helical region and adjacent loops/turns.
|-
| '''7K39, head-binding Fab''' || 14 || 1 || '''15''' || Mainly on HA1 in the head, distributed among accessible loops and neighboring structural elements.
|}

'''Figure placeholders:''' Upload the two PyMOL images to MediaWiki before enabling the lines below.


'''Why the previous 4FQI counts differed:''' The earlier PyMOL screenshot highlighted 18 residues (4 in A and 14 in B). A complete calculation with both antibody protein chains identifies another five within 5 Å: '''A:38, A:40, A:41, A:42 and B:52''', producing 23. An earlier estimate of 25 mistakenly allowed water atoms from the antibody-chain records to contribute contacts, adding A:18 and B:53. The protein-only, solvent-excluded result is '''23'''.

To reproduce the selection in PyMOL, for 4FQI:
<pre>
fetch 4fqi
remove solvent
select ag, (chain A+B) and polymer.protein
select ab, (chain H+L) and polymer.protein
select epitope, byres (ag within 5 of ab)
color red, epitope
</pre>
For 7K39, replace the Fab selection with <code>(chain G+H) and polymer.protein</code> and keep A/B as the chosen antigen protomer.

=== Q3. Which residues form the epitopes? Are they linear? Do they have a linear core? ===

'''4FQI (23 contacts):'''
* HA1, '''A''': 38, 40, 41, 42, 291, 292, 293, 318.
* HA2, '''B''': 18, 19, 20, 21, 36, 37, 38, 41, 42, 45, 46, 48, 49, 52, 56.

'''7K39 (15 contacts):'''
* HA1, '''A''': 48, 49, 50, 90, 91, 92, 93, 269, 270, 271, 272, 273, 284, 285.
* HA2, '''B''': 69.

'''Interpretation:''' Both are predominantly '''conformational (discontinuous) epitopes''', because the contacting residues occur in sequence-distant regions that come together in three dimensions. They contain short contiguous stretches or potential ''linear cores'', for example 4FQI B:18–21 and B:36–38, and 7K39 A:90–93 and A:269–273. These patches do not make the overall epitopes linear or show that the corresponding isolated peptides would bind the antibody.

The complexes do '''not''' target the same region: CR9114 binds the '''stem''' (involving HA1 and HA2), while the 7K39 antibody binds primarily the '''head''' (mostly HA1). HA1=head and HA2=stem are useful teaching approximations, but HA1 also contributes to the structural stem.

== Part B. Sequence-based prediction: BepiPred-2.0 and BepiPred-3.0 ==

'''Data limitation:''' The provided BepiPred-2.0 and BepiPred-3.0 CSV files contain predictions for the '''7K39 H3N2 HA sequences'''. They were not run on the native '''4FQI H5N1''' HA sequences. The two structures contain homologous but non-identical antigens. The quantitative answers in Q4–Q5 therefore directly test the '''7K39 antibody footprint only'''. To evaluate 4FQI independently, run both predictors on the 4FQI FASTA sequences.

The supplied prediction scores were thresholded at '''0.50''' (BepiPred-2.0) and '''0.1512''' (BepiPred-3.0 default, unsmoothed).

=== Q4. Do BepiPred-2.0 predictions overlap the actual epitopes? ===

'''Yes for 7K39, but the result is not very selective.''' All '''15/15''' experimentally observed 7K39 residues are predicted as epitope, giving '''100% recall'''. However, '''407/490''' evaluated residues ('''83.1%''') are predicted as positive, of which 392 are outside the particular observed antibody footprint. Precision is only '''15/407 = 3.69%'''.

A BepiPred-2.0 score of 0.5 therefore recovers the experimental footprint by labeling nearly the whole antigen as a potential epitope. This is not good evidence that it can pinpoint where the specific 7K39 antibody binds. Its score-ranked '''average precision (AP) is 0.0306''', approximately the random-ranking baseline '''15/490 = 0.0306'''.

=== Q5. Does the BepiPred-3.0 prediction overlap now? ===

'''Only slightly for 7K39 at the default threshold.''' Using the ''unsmoothed'' BepiPred-3.0 output, '''2/15''' experimental contact residues are predicted ('''13.3% recall'''): '''A:92''' and '''A:285'''. Their corresponding BepiPred sequence positions are 108 and 301. There are '''128/490''' predicted positive residues (26.1% coverage); precision is '''2/128 = 1.56%''', and '''AP = 0.0297'''.

The optionally ''smoothed'' BepiPred-3.0 score detects '''0/15''' experimental contacts at the same numerical threshold and labels 88/490 residues positive. Smoothing may emphasize contiguous patches but is not guaranteed to improve overlap with a discontinuous antibody footprint.

'''Conclusion for Q4–Q5:''' BepiPred-2.0 has higher recall at its default threshold but very poor localization, while BepiPred-3.0 makes fewer predictions but misses most of this particular antibody footprint. Neither is demonstrably good at ranking the 7K39 contact residues.

=== Q6. What is different in BepiPred-2.0 compared with BepiPred-3.0? ===

{| class="wikitable"
! Feature !! BepiPred-2.0 !! BepiPred-3.0
|-
| Core method || Random Forest using sequence-derived features, trained with structural epitope annotations || Neural networks using protein language-model (ESM-2) representations, trained with updated structural epitope annotations
|-
| Input || Protein sequence || Protein sequence
|-
| Output focus || Mainly continuous/linear epitope prediction with smoothing || Residue-level prediction for potential linear and discontinuous epitope residues, with optional smoothing for linear epitope analysis
|-
| Default threshold used here || 0.50 || 0.1512
|}

BepiPred-3.0 captures richer sequence context using learned embeddings, but neither tool receives the identity or structure of the specific antibody. Predictions are thus potential general B-cell epitope residues, not antibody-specific docking poses. The BepiPred-3.0 structural training annotation uses an antibody-contact cutoff of 4 Å, whereas this exercise uses 5 Å; this difference can also affect observed overlap.

See [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0] and [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0].

== Part C. Structure-based prediction: DiscoTope-2.0 and DiscoTope-3.0 ==

'''Data:''' The uploaded ''Discotope2pred.txt'' and ''file_A_discotope3.csv'' / ''file_B_discotope3.csv'' correspond to the '''7K39 antigen A/B'''. At the supplied settings, DiscoTope-2.0 calls positive residues with combined score greater than approximately '''−3.7'''; DiscoTope-3.0 uses its '''calibrated score ≥ 0.90''' (moderate confidence). The numerical scales are different and cannot be compared directly.

'''Input caveat:''' The DiscoTope-2.0 output includes other HA chains and Fab chains, indicating that the uploaded structure was not strictly antigen-only. DiscoTope-3.0 outputs were reported separately for HA chains A and B. The comparison is therefore useful for the supplied results, but not a fully controlled benchmark of the methods.

=== Q7. Does DiscoTope-2.0 overlap with the actual epitope? ===

'''No, not at the selected threshold for 7K39.''' DiscoTope-2.0 classifies '''58/490''' antigen residues as epitopes (34 in HA1 chain A; 24 in HA2 chain B). '''None of the 15''' experimentally observed 7K39 contact residues is predicted positive, giving '''recall = 0%''' and '''precision = 0%'''. The ranking-based '''AP = 0.0274''', slightly below the 0.0306 random-ranking baseline.

These findings are specific to the supplied output and settings; the method should ideally be rerun on the identical antigen-only structure supplied to the newer version.

=== Q8. Compare the results obtained using DiscoTope-2.0 and DiscoTope-3.0 ===

{| class="wikitable"
! Measurement for 7K39 HA A/B !! DiscoTope-2.0 !! DiscoTope-3.0
|-
| Model concept || Spatial epitope propensities plus residue exposure and neighborhood information || Learned structural representations, including inverse-folding features and exposure information
|-
| Threshold used || Combined score > −3.7 || Calibrated score ≥ 0.90
|-
| Residues predicted positive || 58/490 || 77/490
|-
| Experimental contact residues recovered || 0/15 || '''4/15'''
|-
| Recall || 0% || '''26.7%'''
|-
| Precision || 0% || '''5.19%'''
|-
| Average precision (AP) || 0.0274 || '''0.0466'''
|}

DiscoTope-3.0 correctly identifies experimental contacts '''A:91, A:93, A:271 and B:69'''. At the supplied cutoffs, the two versions share only '''12 predicted-positive HA residues'''. DiscoTope-3.0 therefore agrees better with the known 7K39 footprint in the supplied results, although it still misses 11 of the 15 contacts and predicts 73 residues outside this one interface.

See [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0] and [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0].

=== Q9. Which prediction best matches the experimentally observed epitope? ===

To answer this, compare both '''recall''' and '''precision''', rather than declaring any overlapping prediction correct.

<pre>
Precision = TP / (TP + FP)
Recall = TP / (TP + FN)
</pre>

Here TP denotes predicted residues in the measured 5 Å antibody footprint; FP denotes predicted residues outside that ''specific'' footprint; and FN denotes measured contact residues missed by the method. A threshold-independent score-ranking summary is '''average precision (AP)''', calculated from a precision–recall curve. Its random-ranking baseline here is '''15/490 = 3.06%'''.

{| class="wikitable sortable"
! Method (7K39) !! Correct contacts !! All predicted positives !! Recall !! Precision !! AP !! ROC-AUC
|-
| BepiPred-2.0 || 15/15 || 407/490 || '''100%''' || 3.69% || 0.0306 || 0.492
|-
| BepiPred-3.0, unsmoothed || 2/15 || 128/490 || 13.3% || 1.56% || 0.0297 || 0.492
|-
| DiscoTope-2.0 || 0/15 || 58/490 || 0% || 0% || 0.0274 || 0.413
|-
| '''DiscoTope-3.0''' || '''4/15''' || '''77/490''' || 26.7% || '''5.19%''' || '''0.0466''' || '''0.626'''
|}

'''Best result for this experimental interface: DiscoTope-3.0''', if “best” means highest precision and AP in the supplied data. '''BepiPred-2.0 has the highest recall''', but almost no ability to discriminate the actual interface from the rest of HA at its default threshold. All AP values remain low in absolute terms. The table is not a general benchmark because only one measured antibody footprint has directly matched predictor outputs and the DiscoTope inputs were not identical.



== Part D. Predicting where a known antibody binds (AlphaFold) ==

=== Q10. Are there tools that predict where an isolated antibody binds to an antigen? What do the provided AlphaFold models show? ===

'''Yes.''' Given antigen and antibody sequences, complex-modeling methods such as '''AlphaFold 3''' can propose a three-dimensional antibody–antigen arrangement. Related approaches include '''Boltz''' and docking methods such as '''HADDOCK'''. Unlike BepiPred or DiscoTope, these methods use information about the ''specific antibody'' and aim to model its candidate binding pose. A predicted pose is a hypothesis, not proof of binding.

The provided ZIP archives contain five AlphaFold 3 models each for a hemagglutinin trimer plus antibody chains. They illustrate both successful and problematic complex prediction:

{| class="wikitable"
! Feature !! 4FQI AlphaFold 3 archive !! 7K39 AlphaFold 3 archive
|-
| Experimentally observed epitope || Stem || Head
|-
| Predicted binding location || '''Stem''', consistent across the five models || '''Near the stem''', not the experimental head site; the HA protomer contacted changes in some models
|-
| Example model 0 predicted 5 Å contact residues || 24 (8 in HA1 A, 16 in HA2 B) || 31 (12 in HA1 C, 19 in HA2 D)
|-
| Global complex ipTM, model 0 || 0.85 || 0.62
|-
| Antibody–HA chain-pair ipTM, across supplied models || Approximately '''0.49–0.70''' || Only '''0.08–0.13'''
|-
| Interpretation || Plausible binding pose consistent with the known stem location || Poorly supported antibody–antigen arrangement, despite a moderately high global ipTM value
|}

'''Crucial caution:''' Global ipTM reflects multiple interfaces in a trimeric complex, including HA–HA interfaces. For determining whether the antibody is positioned correctly, examine '''antibody–antigen pairwise ipTM''', interchain predicted aligned error (PAE), and the actual molecular contact geometry. In 7K39, the model-0 minimum antibody–HA pairwise PAE values are about 20–25 Å, indicating considerable uncertainty in the relative placement. This explains why an apparently folded antibody and HA trimer can nevertheless be docked incorrectly.

'''Conclusion:''' The AlphaFold 3 models offer a binding-site hypothesis, and the 4FQI prediction agrees qualitatively with experiment. The 7K39 prediction fails to place the antibody at its experimental head site and has very low antibody–antigen interface confidence. These models should be evaluated by comparison with experimental structures and, where appropriate, binding assays, mutagenesis or other validation.

For confidence-score interpretation see [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 output documentation].

== Overall answer to the exercise ==

The known 4FQI stem and 7K39 head epitopes are spatially discontinuous and are associated with different antibody-binding regions of HA. Sequence-based BepiPred tools can recover candidate epitope residues but, for the supplied 7K39 predictions, do not discriminate the measured antibody footprint well. Of the four submitted predictions, DiscoTope-3.0 gives the strongest agreement with that particular 7K39 footprint (AP 0.0466 versus a random baseline of 0.0306), though the result is still modest. When the identity of an antibody is known, complex prediction can offer a more specific binding pose, but the 7K39 AlphaFold example demonstrates the risk of incorrect docking and the importance of interface confidence metrics. Predictions should be treated as testable hypotheses rather than definitive experimental epitopes.

== Limitations and reproducibility ==

* All experimental epitope counts here use a strict 5 Å '''protein-heavy-atom''' cutoff without solvent; 4FQI = '''23''' and 7K39 = '''15'''.
* BepiPred and DiscoTope numerical performance is directly measurable from the uploaded outputs for '''7K39 only'''. Native-H5 predictions for 4FQI were not supplied, so a proper 4FQI precision–recall evaluation cannot be completed using these outputs.
* The BepiPred residue positions must be translated to PDB numbering (7K39 A: minus 16; 7K39 B: unchanged).
* The DiscoTope-2.0 input includes more chains than the DiscoTope-3.0 antigen-only result, so their numerical comparison is potentially confounded.
* AP is a precision–recall ranking summary; it is not the same quantity as ROC-AUC. The positive class is only 15/490 residues, so conventional classification accuracy is not a useful primary metric.
* Results against only one experimental antibody interface cannot establish that other exposed residues are never epitopes for other antibodies.

== Sources and software ==

* [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction DTU 22145 exercise]
* [https://www.rcsb.org/structure/4FQI PDB 4FQI — H5N1 HA–CR9114]
* [https://www.rcsb.org/structure/7K39 PDB 7K39 — H3N2 HA–head-binding antibody]
* [https://services.healthtech.dtu.dk/services/BepiPred-2.0/ BepiPred-2.0]
* [https://services.healthtech.dtu.dk/services/BepiPred-3.0/ BepiPred-3.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-2.0/ DiscoTope-2.0]
* [https://services.healthtech.dtu.dk/services/DiscoTope-3.0/ DiscoTope-3.0]
* [https://github.com/google-deepmind/alphafold3/blob/main/docs/output.md AlphaFold 3 confidence and output documentation]

Bcell Epitope prediction

2026-10-28T09:56:48Z

Carol:

'''B-cell epitope prediction — Why do some flu vaccines miss?

Influenza A hemagglutinin (HA) has two very different antibody targets: the variable head and the conserved stem. Head-binding antibodies often lose potency as the virus drifts; stem-binding broadly neutralizing antibodies (bnAbs) can resist drift. Can standard sequence-based and structure-based B-cell epitope predictors tell these stories apart?

'''Part A, Get your bearings in PyMOL (find the ground truth epitope)'''

Go to the pdb homepage and look for the pdb entries 4FQI and 7K39. It is hemagglutinin protein from Inlfuenza complexed with two different antibodies. Fetch and inspect structures. You can color the different chains in different colors

* Open 4FQI in PyMOL (fetch 4fqi) and identify chains for HA (HA1/HA2) and Fab (H/L). For ilustrative purposes we will call HA1 the head and HA2 the stem of hemagglutinin.

* Open 7K39 (fetch 7k39) and similarly identify HA vs antibody chains.

hint: you can find this information in the structure web-page, in the PDB file itself (you can open it with a text editor and search for COMPND)

<div style="background-color: lavender; border: solid thin grey;">
:COMPND MOL_ID: 1;
:COMPND 2 MOLECULE: HEMAGGLUTININ HA1;
:COMPND 3 CHAIN: A;
:COMPND 4 FRAGMENT: UNP RESIDUES 17-346;
:COMPND 5 SYNONYM: HEMAGGLUTININ RECEPTOR BINDING SUBUNIT HA1;
:COMPND 6 ENGINEERED: YES;
:COMPND 7 MOL_ID: 2;
:COMPND 8 MOLECULE: HEMAGGLUTININ HA2;
:COMPND 9 CHAIN: B;
:COMPND 10 FRAGMENT: UNP RESIDUES 347-520;
:COMPND 11 SYNONYM: HEMAGGLUTININ MEMBRANE FUSION SUBUNIT HA2;
:COMPND 12 ENGINEERED: YES;
:COMPND 13 MOL_ID: 3;
:COMPND 14 MOLECULE: ANTIBODY CR9114 HEAVY CHAIN;
:COMPND 15 CHAIN: H;
:COMPND 16 FRAGMENT: FAB;
:COMPND 17 ENGINEERED: YES;
:COMPND 18 MOL_ID: 4;
:COMPND 19 MOLECULE: ANTIBODY CR9114 LIGHT CHAIN;
:COMPND 20 CHAIN: L;
:COMPND 21 FRAGMENT: FAB LAMBDA;
:COMPND 22 ENGINEERED: YES
</div>

'''Q1.''' List chain names used in the PDB files for the HA1 and HA2 of the antigen, and the heavy and light antibody chains in each complex.

Define the real epitope from the protein structure complex:

'''Hint''' One of the complexes is actually a trimer, you have several antibody H, L, and antigen chains, it contains a trimer of each. You might want to work with one antibody only as it would be more clear! You can remove the extra chains with the command "remove".
* For each complex, remove waters (remove solvent).
* Create an antibody selection combining the heavy chain and the light chain. (if you do not remember how to do this you can use this [https://teaching.healthtech.dtu.dk/22111/images/0/0b/PyMOL_tutorial_2026.pdf PyMOL tutorial]
* Create an epitope selection on HA as all residues within 5 Å of any antibody atom and color it in red.
'''Hint of PyMOL.''' select epitope, br.(antibody around 5)

'''Q2.''' Provide screenshots of the epitopes for the different protein complexes and the sequence panel with chain letters visible. How many epitope residues are there in head vs stem complexes? Are they mostly loops, helices, or strands (qualitatively)?

'''Q3.''' Can you identify which residues are part of the epitope? Is the epitope linear? Does it have a linear core? Are the two complexes targeting the same chain?

Make antigen-only files

Using pymol, identify the antibody in the structure, select and remove it (action -> remove atoms). Save the resulting molecule as antigen.pdb
If the structure contain several antigen chains that are not part of the epitope, you can remove them.

You can save using the GUI (remember to select PDB as option) as well as typing "save antigen.pdb" in the command line

You can also save the fasta sequence of the antigen with the following command:

save /path/to/folder/XXXX.fasta, all, format=fasta
(remember to change the path and the XXs, name of the PDB complex)

'''Part B. Sequence-based prediction (BepiPred-2.0 & 3.0)'''

Go to the BepiPred2 website and upload the sequences of the stem and the head of hemagglutinin. Leave the default parameters and submit the job. Don't close the window because we will need it again!

'''Q4.''' does the BepiPred predictions overlap with the actual epitopes?

Be careful: pdb file numbering does not always start from 1! Compare the numbering of the pdb with the one returned by BepiPred.

Repeat the prediction now using BepiPred3.

'''Q5.''' Does the prediction overlap with our epitope now?

'''Q6.''' What is different in BepiPred 2 compared to Bepipred 3?

'''Part C. Structure-based prediction (DiscoTope-2.0 & 3.0)'''

Load the antigen.pdb file and submit the prediction in Discotope-2. Save the output as discotope2.pdb

Once again, the prediction is in the b-factor field (but you can download it in other formats as well)
Load it in pymol and colour according to the b-factor.

'''Q7:''' Does the prediction overlap with our epitope?

'''Q8.''' Compare the results obtained in Discotope 2 and 3

'''Q9.''' Which prediction best matches our epitope?



'''Part D. Now suppose that we found a good neutralising antibody against Influenza, we isolate the antibody and have a sequence'''

'''Q10.''' Are there any tools to help us predicting where the antibody will bind in the antigen??? Explore the alpha fold predictions provided in DTU Learn.

Bcell Epitope prediction

2026-10-28T09:52:25Z

Carol:

'''B-cell epitope prediction — Why do some flu vaccines miss?

Influenza A hemagglutinin (HA) has two very different antibody targets: the variable head and the conserved stem. Head-binding antibodies often lose potency as the virus drifts; stem-binding broadly neutralizing antibodies (bnAbs) can resist drift. Can standard sequence-based and structure-based B-cell epitope predictors tell these stories apart?

'''Part A, Get your bearings in PyMOL (find the ground truth epitope)'''

Go to the pdb homepage and look for the pdb entries 4FQI and 7K39. It is hemagglutinin protein from Inlfuenza complexed with two different antibodies. Fetch and inspect structures. You can color the different chains in different colors

* Open 4FQI in PyMOL (fetch 4fqi) and identify chains for HA (HA1/HA2) and Fab (H/L). For ilustrative purposes we will call HA1 the head and HA2 the stem of hemagglutinin.

* Open 7K39 (fetch 7k39) and similarly identify HA vs antibody chains.

hint: you can find this information in the structure web-page, in the PDB file itself (you can open it with a text editor and search for COMPND)

<div style="background-color: lavender; border: solid thin grey;">
:COMPND MOL_ID: 1;
:COMPND 2 MOLECULE: HEMAGGLUTININ HA1;
:COMPND 3 CHAIN: A;
:COMPND 4 FRAGMENT: UNP RESIDUES 17-346;
:COMPND 5 SYNONYM: HEMAGGLUTININ RECEPTOR BINDING SUBUNIT HA1;
:COMPND 6 ENGINEERED: YES;
:COMPND 7 MOL_ID: 2;
:COMPND 8 MOLECULE: HEMAGGLUTININ HA2;
:COMPND 9 CHAIN: B;
:COMPND 10 FRAGMENT: UNP RESIDUES 347-520;
:COMPND 11 SYNONYM: HEMAGGLUTININ MEMBRANE FUSION SUBUNIT HA2;
:COMPND 12 ENGINEERED: YES;
:COMPND 13 MOL_ID: 3;
:COMPND 14 MOLECULE: ANTIBODY CR9114 HEAVY CHAIN;
:COMPND 15 CHAIN: H;
:COMPND 16 FRAGMENT: FAB;
:COMPND 17 ENGINEERED: YES;
:COMPND 18 MOL_ID: 4;
:COMPND 19 MOLECULE: ANTIBODY CR9114 LIGHT CHAIN;
:COMPND 20 CHAIN: L;
:COMPND 21 FRAGMENT: FAB LAMBDA;
:COMPND 22 ENGINEERED: YES
</div>

'''Q1.''' List chain IDs for antigen vs antibody in each complex.

Define the real epitope from the protein structure complex:

'''Hint''' One of the complexes is actually a trimer, you have several antibody H, L, and antigen chains, it contains a trimer of each. You might want to work with one antibody only as it would be more clear! You can remove the extra chains with the command "remove".
* For each complex, remove waters (remove solvent). Create an antibody selection combining the heavy chain and the light chain.
* Create an epitope selection on HA as all residues within 5 Å of any antibody atom and color it in red.
'''Hint of PyMOL.''' select epitope, br.(antibody around 5)

'''Q2.''' Provide screenshots of the epitopes for the different protein complexes and the sequence panel with chain letters visible. How many epitope residues are there in head vs stem complexes? Are they mostly loops, helices, or strands (qualitatively)?

'''Q3.''' Can you identify which residues are part of the epitope? Is the epitope linear? Does it have a linear core? Are the two complexes targeting the same chain?

Make antigen-only files

Using pymol, identify the antibody in the structure, select and remove it (action -> remove atoms). Save the resulting molecule as antigen.pdb
If the structure contain several antigen chains that are not part of the epitope, you can remove them.

You can save using the GUI (remember to select PDB as option) as well as typing "save antigen.pdb" in the command line

You can also save the fasta sequence of the antigen with the following command:

save /path/to/folder/XXXX.fasta, all, format=fasta
(remember to change the path and the XXs, name of the PDB complex)

'''Part B. Sequence-based prediction (BepiPred-2.0 & 3.0)'''

Go to the BepiPred2 website and upload the sequences of the stem and the head of hemagglutinin. Leave the default parameters and submit the job. Don't close the window because we will need it again!

'''Q4.''' does the BepiPred predictions overlap with the actual epitopes?

Be careful: pdb file numbering does not always start from 1! Compare the numbering of the pdb with the one returned by BepiPred.

Repeat the prediction now using BepiPred3.

'''Q5.''' Does the prediction overlap with our epitope now?

'''Q6.''' What is different in BepiPred 2 compared to Bepipred 3?

'''Part C. Structure-based prediction (DiscoTope-2.0 & 3.0)'''

Load the antigen.pdb file and submit the prediction in Discotope-2. Save the output as discotope2.pdb

Once again, the prediction is in the b-factor field (but you can download it in other formats as well)
Load it in pymol and colour according to the b-factor.

'''Q7:''' Does the prediction overlap with our epitope?

'''Q8.''' Compare the results obtained in Discotope 2 and 3

'''Q9.''' Which prediction best matches our epitope?



'''Part D. Now suppose that we found a good neutralising antibody against Influenza, we isolate the antibody and have a sequence'''

'''Q10.''' Are there any tools to help us predicting where the antibody will bind in the antigen??? Explore the alpha fold predictions provided in DTU Learn.

Bcell Epitope prediction

2026-10-28T09:50:29Z

Carol:

'''B-cell epitope prediction — Why do some flu vaccines miss?

Influenza A hemagglutinin (HA) has two very different antibody targets: the variable head and the conserved stem. Head-binding antibodies often lose potency as the virus drifts; stem-binding broadly neutralizing antibodies (bnAbs) can resist drift. Can standard sequence-based and structure-based B-cell epitope predictors tell these stories apart?

'''Part A, Get your bearings in PyMOL (find the ground truth epitope)'''

Go to the pdb homepage and look for the pdb entries 4FQI and 7K39. It is hemagglutinin protein from Inlfuenza complexed with two different antibodies. Fetch and inspect structures. You can color the different chains in different colors

* Open 4FQI in PyMOL (fetch 4fqi) and identify chains for HA (HA1/HA2) and Fab (H/L). For ilustrative purposes we will call HA1 the head and HA2 the stem of hemagglutinin.

* Open 7K39 (fetch 7k39) and similarly identify HA vs antibody chains.

hint: you can find this information in the structure web-page, in the PDB file itself (you can open it with a text editor and search for COMPND)

<div style="background-color: lavender; border: solid thin grey;">
:COMPND MOL_ID: 1;
:COMPND 2 MOLECULE: HEMAGGLUTININ HA1;
:COMPND 3 CHAIN: A;
:COMPND 4 FRAGMENT: UNP RESIDUES 17-346;
:COMPND 5 SYNONYM: HEMAGGLUTININ RECEPTOR BINDING SUBUNIT HA1;
:COMPND 6 ENGINEERED: YES;
:COMPND 7 MOL_ID: 2;
:COMPND 8 MOLECULE: HEMAGGLUTININ HA2;
:COMPND 9 CHAIN: B;
:COMPND 10 FRAGMENT: UNP RESIDUES 347-520;
:COMPND 11 SYNONYM: HEMAGGLUTININ MEMBRANE FUSION SUBUNIT HA2;
:COMPND 12 ENGINEERED: YES;
:COMPND 13 MOL_ID: 3;
:COMPND 14 MOLECULE: ANTIBODY CR9114 HEAVY CHAIN;
:COMPND 15 CHAIN: H;
:COMPND 16 FRAGMENT: FAB;
:COMPND 17 ENGINEERED: YES;
:COMPND 18 MOL_ID: 4;
:COMPND 19 MOLECULE: ANTIBODY CR9114 LIGHT CHAIN;
:COMPND 20 CHAIN: L;
:COMPND 21 FRAGMENT: FAB LAMBDA;
:COMPND 22 ENGINEERED: YES
</div>

'''Q1.''' List chain IDs for antigen vs antibody in each complex.

Define the real epitope from the protein structure complex.
''Hint'' In one of the complexes you have several antibody and antigen chains, it contains a trimer of each. You might want to work with one antibody only as it would be more clear!
You can remove the extra chains with the command "remove".
For each complex, remove waters (remove solvent). Create an antibody selection combining the heavy chain and the light chain.
Create an epitope selection on HA as all residues within 5 Å of any antibody atom and color it in red.
''Hint of PyMOL.'' select epitope, br.(antibody around 5)

'''Q2.''' Provide screenshots of the epitopes for the different protein complexes and the sequence panel with chain letters visible. How many epitope residues are there in head vs stem complexes? Are they mostly loops, helices, or strands (qualitatively)?

'''Q3.''' Can you identify which residues are part of the epitope? Is the epitope linear? Does it have a linear core? Are the two complexes targeting the same chain?

Make antigen-only files

Using pymol, identify the antibody in the structure, select and remove it (action -> remove atoms). Save the resulting molecule as antigen.pdb
If the structure contain several antigen chains that are not part of the epitope, you can remove them.

You can save using the GUI (remember to select PDB as option) as well as typing "save antigen.pdb" in the command line

You can also save the fasta sequence of the antigen with the following command:

save /path/to/folder/XXXX.fasta, all, format=fasta
(remember to change the path and the XXs, name of the PDB complex)

'''Part B. Sequence-based prediction (BepiPred-2.0 & 3.0)'''

Go to the BepiPred2 website and upload the sequences of the stem and the head of hemagglutinin. Leave the default parameters and submit the job. Don't close the window because we will need it again!

'''Q4.''' does the BepiPred predictions overlap with the actual epitopes?

Be careful: pdb file numbering does not always start from 1! Compare the numbering of the pdb with the one returned by BepiPred.

Repeat the prediction now using BepiPred3.

'''Q5.''' Does the prediction overlap with our epitope now?

'''Q6.''' What is different in BepiPred 2 compared to Bepipred 3?

'''Part C. Structure-based prediction (DiscoTope-2.0 & 3.0)'''

Load the antigen.pdb file and submit the prediction in Discotope-2. Save the output as discotope2.pdb

Once again, the prediction is in the b-factor field (but you can download it in other formats as well)
Load it in pymol and colour according to the b-factor.

'''Q7:''' Does the prediction overlap with our epitope?

'''Q8.''' Compare the results obtained in Discotope 2 and 3

'''Q9.''' Which prediction best matches our epitope?



'''Part D. Now suppose that we found a good neutralising antibody against Influenza, we isolate the antibody and have a sequence'''

'''Q10.''' Are there any tools to help us predicting where the antibody will bind in the antigen??? Explore the alpha fold predictions provided in DTU Learn.

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-28T08:58:28Z

Carol: /* Wednesday, 2 September 2026 — Introduction and databases */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AllergenicityPrediction Allergenicity Prediction]

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction Antibody Structure Prediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B Cell Epitope prediction]'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-28T08:58:11Z

Carol: /* Wednesday, 21 October 2026 — BCR and TCR structure prediction */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AllergenicityPrediction Allergenicity Prediction]

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction Antibody Structure Prediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B Cell Epitope prediction]'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-28T08:57:58Z

Carol: /* Wednesday, 28 October 2026 — B-cell epitope prediction */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AllergenicityPrediction Allergenicity Prediction]

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction Antibody Structure Prediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B Cell Epitope prediction]'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-28T08:57:34Z

Carol: /* Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AllergenicityPrediction Allergenicity Prediction]

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction Antibody Structure Prediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B Cell Epitope prediction]

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-28T08:57:14Z

Carol: /* Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises: https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AllergenicityPrediction Allergenicity Prediction'''

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction Antibody Structure Prediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B Cell Epitope prediction]

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-28T08:55:31Z

Carol: /* Wednesday, 21 October 2026 — BCR and TCR structure prediction */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises: To be updated'''

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction Antibody Structure Prediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B Cell Epitope prediction]

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-28T08:55:04Z

Carol: /* Wednesday, 28 October 2026 — B-cell epitope prediction */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises: To be updated'''

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction AntibodyStructurePrediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B Cell Epitope prediction]

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-28T08:54:12Z

Carol: /* Wednesday, 28 October 2026 — B-cell epitope prediction */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises: To be updated'''

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction AntibodyStructurePrediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise:''' [https://teaching.healthtech.dtu.dk/22145/index.php/Bcell_Epitope_prediction B cell Epitope prediction]

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-28T08:01:09Z

Carol: /* Wednesday, 11 November 2026 — Written EXAM */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises: To be updated'''

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction AntibodyStructurePrediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise:'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 8.45 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

Bcell Epitope prediction

2026-10-27T16:45:16Z

Carol:

'''B-cell epitope prediction — Why do some flu vaccines miss?

Influenza A hemagglutinin (HA) has two very different antibody targets: the variable head and the conserved stem. Head-binding antibodies often lose potency as the virus drifts; stem-binding broadly neutralizing antibodies (bnAbs) can resist drift. Can standard sequence-based and structure-based B-cell epitope predictors tell these stories apart?

'''Part A, Get your bearings in PyMOL (find the ground truth epitope)'''

Go to the pdb homepage and look for the pdb entries 4FQI and 7K39. It is hemagglutinin protein from Inlfuenza complexed with two different antibodies. Fetch and inspect structures. You can color the different chains in different colors

* Open 4FQI in PyMOL (fetch 4fqi) and identify chains for HA (HA1/HA2) and Fab (H/L). For ilustrative purposes we will call HA1 the head and HA2 the stem of hemagglutinin.

* Open 7K39 (fetch 7k39) and similarly identify HA vs antibody chains.

hint: you can find this information in the structure web-page, in the PDB file itself (you can open it with a text editor and search for COMPND)

<div style="background-color: lavender; border: solid thin grey;">
:COMPND MOL_ID: 1;
:COMPND 2 MOLECULE: HEMAGGLUTININ HA1;
:COMPND 3 CHAIN: A;
:COMPND 4 FRAGMENT: UNP RESIDUES 17-346;
:COMPND 5 SYNONYM: HEMAGGLUTININ RECEPTOR BINDING SUBUNIT HA1;
:COMPND 6 ENGINEERED: YES;
:COMPND 7 MOL_ID: 2;
:COMPND 8 MOLECULE: HEMAGGLUTININ HA2;
:COMPND 9 CHAIN: B;
:COMPND 10 FRAGMENT: UNP RESIDUES 347-520;
:COMPND 11 SYNONYM: HEMAGGLUTININ MEMBRANE FUSION SUBUNIT HA2;
:COMPND 12 ENGINEERED: YES;
:COMPND 13 MOL_ID: 3;
:COMPND 14 MOLECULE: ANTIBODY CR9114 HEAVY CHAIN;
:COMPND 15 CHAIN: H;
:COMPND 16 FRAGMENT: FAB;
:COMPND 17 ENGINEERED: YES;
:COMPND 18 MOL_ID: 4;
:COMPND 19 MOLECULE: ANTIBODY CR9114 LIGHT CHAIN;
:COMPND 20 CHAIN: L;
:COMPND 21 FRAGMENT: FAB LAMBDA;
:COMPND 22 ENGINEERED: YES
</div>

'''Q1.''' List chain IDs for antigen vs antibody in each complex.

Define the real epitope. For each complex, remove waters (remove solvent). Create an epitope selection on HA as all residues within 5 Å of any antibody atom and color it in red.

''Hint of PyMOL.'' select epitope, br.(antibody around 5)
''Hint2'' In one of the complexes you have several antibody H and L chains. You might want to work with one antibody only!

'''Q2.''' Provide screenshots of the epitopes for the different protein complexes and the sequence panel with chain letters visible. How many epitope residues are there in head vs stem complexes? Are they mostly loops, helices, or strands (qualitatively)?

'''Q3.''' Can you identify which residues are part of the epitope? Is the epitope linear? Does it have a linear core? Are the two complexes targeting the same chain?

Make antigen-only files

Using pymol, identify the antibody in the structure, select and remove it (action -> remove atoms). Save the resulting molecule as antigen.pdb
If the structure contain several antigen chains that are not part of the epitope, you can remove them.

You can save using the GUI (remember to select PDB as option) as well as typing "save antigen.pdb" in the command line

You can also save the fasta sequence of the antigen with the following command:

save /path/to/folder/XXXX.fasta, all, format=fasta
(remember to change the path and the XXs, name of the PDB complex)

'''Part B. Sequence-based prediction (BepiPred-2.0 & 3.0)'''

Go to the BepiPred2 website and upload the sequences of the stem and the head of hemagglutinin. Leave the default parameters and submit the job. Don't close the window because we will need it again!

'''Q4.''' does the BepiPred predictions overlap with the actual epitopes?

Be careful: pdb file numbering does not always start from 1! Compare the numbering of the pdb with the one returned by BepiPred.

Repeat the prediction now using BepiPred3.

'''Q5.''' Does the prediction overlap with our epitope now?

'''Q6.''' What is different in BepiPred 2 compared to Bepipred 3?

'''Part C. Structure-based prediction (DiscoTope-2.0 & 3.0)'''

Load the antigen.pdb file and submit the prediction in Discotope-2. Save the output as discotope2.pdb

Once again, the prediction is in the b-factor field (but you can download it in other formats as well)
Load it in pymol and colour according to the b-factor.

'''Q7:''' Does the prediction overlap with our epitope?

'''Q8.''' Compare the results obtained in Discotope 2 and 3

'''Q9.''' Which prediction best matches our epitope?



'''Part D. Now suppose that we found a good neutralising antibody against Influenza, we isolate the antibody and have a sequence'''

'''Q10.''' Are there any tools to help us predicting where the antibody will bind in the antigen??? Explore the alpha fold predictions provided in DTU Learn.

Bcell Epitope prediction

2026-10-27T16:44:55Z

Carol:

'''B-cell epitope prediction — Why do some flu vaccines miss?

Influenza A hemagglutinin (HA) has two very different antibody targets: the variable head and the conserved stem. Head-binding antibodies often lose potency as the virus drifts; stem-binding broadly neutralizing antibodies (bnAbs) can resist drift. Can standard sequence-based and structure-based B-cell epitope predictors tell these stories apart?

'''Part A, Get your bearings in PyMOL (find the ground truth epitope)'''

Go to the pdb homepage and look for the pdb entries 4FQI and 7K39. It is hemagglutinin protein from Inlfuenza complexed with two different antibodies. Fetch and inspect structures. You can color the different chains in different colors

* Open 4FQI in PyMOL (fetch 4fqi) and identify chains for HA (HA1/HA2) and Fab (H/L). For ilustrative purposes we will call HA1 the head and HA2 the stem of hemagglutinin.

* Open 7K39 (fetch 7k39) and similarly identify HA vs antibody chains.

hint: you can find this information in the structure web-page, in the PDB file itself (you can open it with a text editor and search for COMPND)

<div style="background-color: lavender; border: solid thin grey;">
:COMPND MOL_ID: 1;
:COMPND 2 MOLECULE: HEMAGGLUTININ HA1;
:COMPND 3 CHAIN: A;
:COMPND 4 FRAGMENT: UNP RESIDUES 17-346;
:COMPND 5 SYNONYM: HEMAGGLUTININ RECEPTOR BINDING SUBUNIT HA1;
:COMPND 6 ENGINEERED: YES;
:COMPND 7 MOL_ID: 2;
:COMPND 8 MOLECULE: HEMAGGLUTININ HA2;
:COMPND 9 CHAIN: B;
:COMPND 10 FRAGMENT: UNP RESIDUES 347-520;
:COMPND 11 SYNONYM: HEMAGGLUTININ MEMBRANE FUSION SUBUNIT HA2;
:COMPND 12 ENGINEERED: YES;
:COMPND 13 MOL_ID: 3;
:COMPND 14 MOLECULE: ANTIBODY CR9114 HEAVY CHAIN;
:COMPND 15 CHAIN: H;
:COMPND 16 FRAGMENT: FAB;
:COMPND 17 ENGINEERED: YES;
:COMPND 18 MOL_ID: 4;
:COMPND 19 MOLECULE: ANTIBODY CR9114 LIGHT CHAIN;
:COMPND 20 CHAIN: L;
:COMPND 21 FRAGMENT: FAB LAMBDA;
:COMPND 22 ENGINEERED: YES
</div>

'''Q1.''' List chain IDs for antigen vs antibody in each complex.

Define the real epitope. For each complex, remove waters (remove solvent). Create an epitope selection on HA as all residues within 5 Å of any antibody atom and color it in red.

''Hint of PyMOL.'' select epitope, br.(antibody around 5)
''Hint2'' In one of the complexes you have several antibody H and L chains. You might want to work with one antibody only!

'''Q2.''' Provide screenshots of the epitopes for the different protein complexes and the sequence panel with chain letters visible. How many epitope residues are there in head vs stem complexes? Are they mostly loops, helices, or strands (qualitatively)?

'''Q3.''' Can you identify which residues are part of the epitope? Is the epitope linear? Does it have a linear core? Are the two complexes targeting the same chain?

Make antigen-only files

Using pymol, identify the antibody in the structure, select and remove it (action -> remove atoms). Save the resulting molecule as antigen.pdb
If the structure contain several antigen chains that are not part of the epitope, you can remove them.

You can save using the GUI (remember to select PDB as option) as well as typing "save antigen.pdb" in the command line

You can also save the fasta sequence of the antigen with the following command:

save /path/to/folder/XXXX.fasta, all, format=fasta
(remember to change the path and the XXs, name of the PDB complex)

'''Part B. Sequence-based prediction (BepiPred-2.0 & 3.0)'''

Go to the BepiPred2 website and upload the sequences of the stem and the head of hemagglutinin. Leave the default parameters and submit the job. Don't close the window because we will need it again!

'''Q4.''' does the BepiPred predictions overlap with the actual epitopes?

Be careful: pdb file numbering does not always start from 1! Compare the numbering of the pdb with the one returned by BepiPred.

Repeat the prediction now using BepiPred3.

'''Q5.''' Does the prediction overlap with our epitope now?

'''Q6.''' What is different in BepiPred 2 compared to Bepipred 3?

'''Part C. Structure-based prediction (DiscoTope-2.0 & 3.0)'''

Load the antigen.pdb file and submit the prediction in Discotope-2. Save the output as discotope2.pdb

Once again, the prediction is in the b-factor field (but you can download it in other formats as well)
Load it in pymol and colour according to the b-factor.

'''Q7:''' Does the prediction overlap with our epitope?

'''Q8.''' Compare the results obtained in Discotope 2 and 3

'''Q9.''' Which prediction best matches our epitope?



'''Q11.''' Are there any tools to help us predicting where the antibody will bind in the antigen??? Explore the alpha fold predictions provided in DTU Learn.

Bcell Epitope prediction

2026-10-27T16:44:09Z

Carol:

'''B-cell epitope prediction — Why do some flu vaccines miss?

Influenza A hemagglutinin (HA) has two very different antibody targets: the variable head and the conserved stem. Head-binding antibodies often lose potency as the virus drifts; stem-binding broadly neutralizing antibodies (bnAbs) can resist drift. Can standard sequence-based and structure-based B-cell epitope predictors tell these stories apart?

'''Part A, Get your bearings in PyMOL (find the ground truth epitope)'''

Go to the pdb homepage and look for the pdb entries 4FQI and 7K39. It is hemagglutinin protein from Inlfuenza complexed with two different antibodies. Fetch and inspect structures. You can color the different chains in different colors

* Open 4FQI in PyMOL (fetch 4fqi) and identify chains for HA (HA1/HA2) and Fab (H/L). For ilustrative purposes we will call HA1 the head and HA2 the stem of hemagglutinin.

* Open 7K39 (fetch 7k39) and similarly identify HA vs antibody chains.

hint: you can find this information in the structure web-page, in the PDB file itself (you can open it with a text editor and search for COMPND)

<div style="background-color: lavender; border: solid thin grey;">
:COMPND MOL_ID: 1;
:COMPND 2 MOLECULE: HEMAGGLUTININ HA1;
:COMPND 3 CHAIN: A;
:COMPND 4 FRAGMENT: UNP RESIDUES 17-346;
:COMPND 5 SYNONYM: HEMAGGLUTININ RECEPTOR BINDING SUBUNIT HA1;
:COMPND 6 ENGINEERED: YES;
:COMPND 7 MOL_ID: 2;
:COMPND 8 MOLECULE: HEMAGGLUTININ HA2;
:COMPND 9 CHAIN: B;
:COMPND 10 FRAGMENT: UNP RESIDUES 347-520;
:COMPND 11 SYNONYM: HEMAGGLUTININ MEMBRANE FUSION SUBUNIT HA2;
:COMPND 12 ENGINEERED: YES;
:COMPND 13 MOL_ID: 3;
:COMPND 14 MOLECULE: ANTIBODY CR9114 HEAVY CHAIN;
:COMPND 15 CHAIN: H;
:COMPND 16 FRAGMENT: FAB;
:COMPND 17 ENGINEERED: YES;
:COMPND 18 MOL_ID: 4;
:COMPND 19 MOLECULE: ANTIBODY CR9114 LIGHT CHAIN;
:COMPND 20 CHAIN: L;
:COMPND 21 FRAGMENT: FAB LAMBDA;
:COMPND 22 ENGINEERED: YES
</div>

'''Q1.''' List chain IDs for antigen vs antibody in each complex.

Define the real epitope. For each complex, remove waters (remove solvent). Create an epitope selection on HA as all residues within 5 Å of any antibody atom and color it in red.

''Hint of PyMOL.'' select epitope, br.(antibody around 5)
''Hint2'' In one of the complexes you have several antibody H and L chains. You might want to work with one antibody only!

'''Q2.''' Provide screenshots of the epitopes for the different protein complexes and the sequence panel with chain letters visible. How many epitope residues are there in head vs stem complexes? Are they mostly loops, helices, or strands (qualitatively)?

'''Q3.''' Can you identify which residues are part of the epitope? Is the epitope linear? Does it have a linear core? Are the two complexes targeting the same chain?

Make antigen-only files

Using pymol, identify the antibody in the structure, select and remove it (action -> remove atoms). Save the resulting molecule as antigen.pdb
If the structure contain several antigen chains that are not part of the epitope, you can remove them.

You can save using the GUI (remember to select PDB as option) as well as typing "save antigen.pdb" in the command line

You can also save the fasta sequence of the antigen with the following command:

save /path/to/folder/XXXX.fasta, all, format=fasta
(remember to change the path and the XXs, name of the PDB complex)

'''Part B. Sequence-based prediction (BepiPred-2.0 & 3.0)'''

Go to the BepiPred2 website and upload the sequences of the stem and the head of hemagglutinin. Leave the default parameters and submit the job. Don't close the window because we will need it again!

'''Q4.''' does the BepiPred predictions overlap with the actual epitopes?

Be careful: pdb file numbering does not always start from 1! Compare the numbering of the pdb with the one returned by BepiPred.

Repeat the prediction now using BepiPred3.

'''Q5.''' Does the prediction overlap with our epitope now?

'''Q6.''' What is different in BepiPred 2 compared to Bepipred 3?

'''Part C. Structure-based prediction (DiscoTope-2.0 & 3.0)'''

Load the antigen.pdb file and submit the prediction in Discotope-2. Save the output as discotope2.pdb

Once again, the prediction is in the b-factor field (but you can download it in other formats as well)
Load it in pymol and colour according to the b-factor.

'''Q7:''' Does the prediction overlap with our epitope?

'''Q8.''' Compare the results obtained in Discotope 2 and 3

'''Q9.''' Which prediction best matches our epitope?

<--'''Part D. Antigenic drift vs conservation'''

Compare another HA
Fetch an HA from a different subtype/strain (e.g., another H5N1, H1N1, or the C05-bound head example). For convenience, you can use 4FQI as reference.

Align HA chains (align) and map your epitope residues from the first antigen onto the second by position.

'''Q10.''' Which epitope class (head vs stem) is more conserved by sequence identity at the interface?

'''Part D. Now suppose that we found a good neutralising antibody against Influenza, we isolate the antibody and have a sequence'''
-->

'''Q11.''' Are there any tools to help us predicting where the antibody will bind in the antigen??? Explore the alpha fold predictions provided in DTU Learn.

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-21T08:41:29Z

Carol: /* Wednesday, 11 November 2026 — Written EXAM */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises: To be updated'''

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction AntibodyStructurePrediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise:'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 9-11 '''EXAM'''
: 11:40 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-21T08:40:32Z

Carol: /* Wednesday, 28 October 2026 — B-cell epitope prediction */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises: To be updated'''

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction AntibodyStructurePrediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — Paolo Marcatili, Associate Director, in silico biologics discovery, Novo Nordisk
:'''Slides:''' Available on DTU Learn.
:'''Exercise:'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 9-11 '''EXAM'''
: 11-12 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-21T08:39:25Z

Carol: /* Wednesday, 21 October 2026 — BCR and TCR structure prediction */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises: To be updated'''

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction AntibodyStructurePrediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — TBD.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 9-11 '''EXAM'''
: 11-12 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

22145 Immunological Bioinformatics Course Programme Fall 2026

2026-10-21T08:38:58Z

Carol: /* Lecture and exercise plan */

== General information ==

=== Where and when ===

Classes are held on Wednesdays from 8:00 to 12:00 in '''Building 208, Holdlokale 3/065''', beginning on '''Wednesday, 2 September 2026, at 09:00'''.

The first hour of each class 8-9 is dedicated to course preparation, including videos, papers, or other reading material provided on DTU Learn. Unless otherwise specified, we meet at the classroom at 09:00.

=== Teachers ===

* [https://www.dtu.dk/Person/cwis?id=142840&entity=profile Carolina Barra Quaglia] — Associate Professor and course responsible.
* [https://www.dtu.dk/person/morten-nielsen?id=5973&entity=profile Morten Nielsen] — Professor and course responsible.

=== Teaching assistant ===

* [https://orbit.dtu.dk/en/persons/eric-bautista-farrerons/ Eric Bautista Farrerons] — PhD student.

=== Course content ===

The course introduces students to state-of-the-art methods in computational immunology.

The methods are presented in the context of immunology as a domain-specific field. Introductions to the underlying theory are followed by practical exercises, enabling students to perform analyses independently. The course covers immunological bioinformatics and computational vaccinology, with perspectives on infectious diseases, cancer immunotherapy, and autoimmunity.

The course is taught in two parts:
* Part 1 consists of lectures and group-based exercises. Part 1 ends with a "in class" written exam on November 11th.
* Part 2 consists of a group research project, culminating in a poster presentation on the final day of the course. December 2nd.

See also [https://kurser.dtu.dk/course/2026-2027/22145 the DTU course description for course 22145].

=== Curriculum ===

There is no required textbook. The curriculum consists of lectures and exercises completed in class, supplemented by papers and book chapters made available on DTU Learn. Please note that ''all'' exercise guides are part of the mandatory curriculum, including the ''answers'', which will be made available on DTU Learn after each exercise.

=== Hand-ins ===

Regardless of the software used to prepare your work, '''all submissions must be handed in as PDF files'''.

Group submissions are permitted and recommended.

''The hand-ins are mandatory however, they do not affect your grade.'' Their main purpose is to help you prepare for the exam and to help the teaching team assess students' understanding. If several participants make the same mistake, the issue will be clarified at the beginning of the following lecture.

=== Exam ===

The assessment consists of two parts: a written two-hour written exam in week 10, and a group oral project presentation with individual questions at the end of the course. The final grade is based on an overall assessment of both parts. To pass the course, you must complete the oral presentation and pass the written exam.

All aids are permitted during the written exam, including books, papers, and notes.
'''Internet is not allowed in the written exam'''
You are ''not'' permitted to communicate with other people through email, social media, chat services, or any file-sharing platforms. Internet traffic may be logged during the exam to ensure compliance with these restrictions.

The use of generative AI is permitted during the project work, but you must clearly state where and how you used them as stated in [https://www.bibliotek.dtu.dk/en/research-service/reference-management/kunstig-intelligens DTU guidelines].

Please beware that content, results, links etc. that you discover via generative AI may be incorrect - simply because of the way generative AI models are built.

Generative AI produces content from huge amounts of data collected from the internet. The data that the AI system is based on is not necessarily publicly available and it may be copyright protected.

You are solely responsible to adhere to the academic standards of honesty, transparency and accountability as outlined in DTU's code of honour and [https://www.inside.dtu.dk/en/research/good-scientific-conduct Principles for good scientific conduct].

=== DTU Learn link ===

* DTU Learn page: https://learn.inside.dtu.dk/d2l/home/325135

== Lecture and exercise plan ==

'''Note:''' This is a ''preliminary'' plan and may be subject to changes. Check the page regularly for updates.

=== Wednesday, 2 September 2026 — Introduction and databases ===

:'''Preparation material:'''
:* ''Video: Immunology Recap''
:* ''Video: Adaptive Immunity''
:* ''Book chapter: The adaptive Immune system''

:'''Lectures: We start at 9am'''
:* ''Introduction to immunological bioinformatics'' — Carolina Barra Quaglia.
:* ''Databases'' — Carolina Barra Quaglia.
:'''Slides:''' All slides will be made available on DTU Learn.

:'''Exercise:[https://teaching.healthtech.dtu.dk/22145/index.php/IEDB IEDB]'''

=== Wednesday, 9 September 2026 — MHC binding prediction, Part 1 ===

:'''Lecture:''' ''MHC binding prediction, Part 1'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part1.php MHC binding prediction]'''

=== Wednesday, 16 September 2026 — MHC binding prediction, Part 2 ===

:'''Lecture:''' ''MHC binding prediction, Part 2'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_MHC_bind/exercise_2023_part2.php MHC Binding Prediction 2]'''


=== Wednesday, 23 September 2026 — Antigen processing and presentation ===

:'''Lecture:''' ''Antigen processing and presentation'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise: [https://teaching.healthtech.dtu.dk/22145/index.php?title=Exercise:AntigenProcessing Antigen Processing]'''

=== Wednesday, 30 September 2026 — Allergenicity and immunogenicity prediction ===

:'''Lectures:'''
:* ''Allergenicity prediction'' — Carolina Barra Quaglia.
:* ''Immunogenicity prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercises: To be updated'''

=== Wednesday, 7 October 2026 — Vaccine design ===

:'''Lecture:''' ''Vaccine design'' — Morten Nielsen.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:[https://teaching.healthtech.dtu.dk/morten_teaching/27485.imm/exercise_HCVVaccine/exercise_HCVVaccine_2023.php HCV Vaccine development]'''

=== Wednesday, 14 October 2026 — Autumn break ===

=== Wednesday, 21 October 2026 — BCR and TCR structure prediction ===

:'''Workshop:''' ''BCR and TCR structure prediction'' — Carolina Barra Quaglia.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:'''[https://teaching.healthtech.dtu.dk/22145/index.php/Exercise:AntibodyStructurePrediction AntibodyStructurePrediction]'''

=== Wednesday, 28 October 2026 — B-cell epitope prediction ===

:'''Lecture:''' ''B-cell epitope prediction'' — TBD.
:'''Slides:''' Available on DTU Learn.
:'''Exercise:'''

=== Wednesday, 4 November 2026 — T-Cell Receptor interaction predictions ===

:'''Lecture:''' ''TCR interaction models'' — Morten Nielsen
:'''Slides:''' Available on DTU Learn.
:'''Exercises:'''


=== Wednesday, 11 November 2026 — Written EXAM ===

: 9-11 '''EXAM'''
: 11-12 '''Introduction:''' ''Short introduction to project work'' — Carolina Barra Quaglia.

=== Wednesday, 18 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 25 November 2026 — Project work ===

:'''Lecture:''' No lecture.

=== Wednesday, 2 December 2026 — Poster session ===

:Poster presentations will begin at '''08:00'''.

:'''Poster presentation agenda:''' To be announced.

== Exam ==

In order to get access to the written exam you should submit first all the exercises

The exam has two parts: A 2hs written exam on '''November 11th''' and an oral presentation of the group project on '''December 2nd'''. Failure to pass one of the two will correspond to the No Pass grade

Exercise:AntibodyStructurePrediction

2026-10-21T03:17:08Z

Carol: /* Identifying mutations in the antibody sequence */

= Antibody Structure Modelling: From Sequence to 3D Structure =

__TOC__

== Introduction: ==

Antibodies recognize antigens through their variable domains, which contain six complementarity-determining regions (CDRs): three in the heavy chain and three in the light chain. These loops form the antigen-binding site and are supported by more conserved framework (FW) regions.

Predicting the three-dimensional structure of antibodies helps us understand how their sequence, shape, and chemical properties may influence antigen recognition. Different computational approaches, from traditional template-based modelling to modern deep-learning methods, can be used to predict antibody structures.

In this exercise, you will model an antibody from a patient with chronic lymphocytic leukemia (CLL), explore its structure using PyMOL, and investigate how sequence variation and structural features may influence antigen binding.

== Learning objectives ==

By the end of this exercise, you should be able to:

* Identify the framework and CDR regions of an antibody.
* Understand the basic principles of template-based antibody modelling.
* Generate and visualize a three-dimensional antibody model using computational tools.
* Explore how CDR sequence, structure, and electrostatic properties may influence antigen recognition.
* Identify somatic mutations and discuss their potential effects on antibody binding.
* Compare antibody structures predicted using traditional template-based and modern deep-learning approaches.

== Antibody sequences ==

We will use the variable regions of the SI135 antibody.

'''Light chain (VL):'''

<pre>
>SI135_VL
SYVVTQPPSVSVAPGKTARITCGGNNIGSKSVHWYQQKPGQAPVLVIYYDNDRPSGIPERFSGSNSGNTATLTISRVEAGDEADYYCQVWDSGSDHPWVFGGGTKLTVL
</pre>

'''Heavy chain (VH):'''

<pre>
>SI135_VH
EVQLVESGGGLVRPGGSLRLSCAASGFTFSTYSMNWVRQAPGKGLEWVSSISSSSNSIYCADSVKGRFTISRDNANNSLYMQMNSLRAEDTAVYYCARDVNTMDVWGQGILVTVLL
</pre>

== Part 1 — Identify the antibody regions ==

Go to the [https://services.healthtech.dtu.dk/services/LYRA-1.0/ Lyra] web page.

''Lyra is a tool for predicting antibody and TCR structures. It is possible to use Lyra in automated or expert mode.''

Insert the light and heavy chains in the proper text area and submit. After a few seconds you will arrive to the output page, where the alignment and all the templates are displayed.

Note that for each Region of each chain of the antibody a Template has been found.

{| class="wikitable"
! Region
! Templates
|-
| Framework
| ...
|-
| Loop 1 (CDR1)
| ...
|-
| Loop 2 (CDR2)
| ...
|-
| Loop 3 (CDR3)
| ...
|}

Note that after the Regions there is an extra template used for packing the whole structure.

Explore the output page and the different tabs - check the PDB file, Log, and manual selection ones to find this info.

Now answer the following questions:

'''Q1: what is the sequence similarity of the best H and L templates for the framework (FW)?'''

'''Q2: What is the light chain type? Is it the most common in humans?'''

=== Sequence alignment and CDR modelling ===

Focus on the sequence alignment used to generate the model. You can find it in one of the tabs in Lyra website, as well as in the PDB file itself.

'''Q3: where are the differences between target and template located? Which are the most drastic that could affect the modeling process (e.g. G->W)?'''

''hint: if you don't remeber the amino acids, you can use the Blosum62 matrix to estimate the expected effect of a substitution: the lower the number, the larger the expected impact. For sake of time, you only need to give a qualitative answer.''

'''Q4: Do the loops/CDR have the same length in target and template? What does it mean, in terms of expected accuracy?'''

=== Selection of structural templates ===

Now, look at the templates automatically selected by Lyra to build the final model.

'''Q5: How many templates were used? Which is the region for which the template is more dissimilar to the target sequence?'''

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Visualizing the antigen-binding site (ABS) ===

Download the pdb file and open it with pymol. There is a small "S" bottom right, it shows/hides the sequence. Show the whole protein as cartoon and color by chain. Select and color in red all the CDRs. (the selected residues appear as the "sele" item in the menu). The CDR definition can be seen in the alignment generated by Lyra.

'''Q6: What is the general shape of the ABS: flat, a pocket, a mountain or a groove?''' ''It can help to show the antibody as surface''

=== CDR loop structure ===

'''Q7: Which loop is the longest and protruding the most? Is this the most common case?'''

== Part 3 — Somatic mutations ==

=== Identifying mutations in the antibody sequence ===

Use the [http://www.imgt.org/3Dstructure-DB/cgi/DomainGapAlign.cgi Domain Gap Align tool] identify the somatic mutations in the V and J regions. You need to use the amino acid sequence and select human as the source organism. In PyMol, show the mutations as sticks (main chain and side chain) and color them green.

Unfortunately today the server has a planning update so the output of the tool are provided below

[[image:Imgt_GapAlign_results_VH.png|1000px|center]]

[[image:Imgt_GapAlign_results_VL.png|1000px|center]]

'''Q8: Which Mutations are located inside/near the CDRs? Which are the ones that are most likely to have an impact on the binding of the antibody?''' ''A qualitative answer is sufficient''

== Part 4 — Electrostatic properties of the antigen-binding site ==

=== Visualizing the electrostatic potential ===

Generate the electrostatic potentials of the Ig ('''A'''ction -> generate -> vacuum electrostatic -> protein contact potentials).

'''Q9: Identify the residues in the CDRs that mostly contribute to the charge distribution.'''

''highlight them in the sequence and look for charged residues such as K, R, D, E''

=== Modifying the electrostatic properties (optional) ===

'''Q10: (optional) Mutate some residues in order to have a positive charge in the middle of the antigen binding site, then recalculate the electrostatic potentials. Can you see the changes?'''

== Part 5 — Comparing antibody modelling approaches ==

=== LYRA versus ABodyBuilder2 ===

'''Q11: Go to the AbodyBuilder 2 site and make a new structure prediction. Download the PDB file and compare with the model generated by LYRA. Can you see some differences?'''

'''Finished!'''

Exercise:AntibodyStructurePrediction

2026-10-21T03:16:53Z

Carol: /* Identifying mutations in the antibody sequence */

= Antibody Structure Modelling: From Sequence to 3D Structure =

__TOC__

== Introduction: ==

Antibodies recognize antigens through their variable domains, which contain six complementarity-determining regions (CDRs): three in the heavy chain and three in the light chain. These loops form the antigen-binding site and are supported by more conserved framework (FW) regions.

Predicting the three-dimensional structure of antibodies helps us understand how their sequence, shape, and chemical properties may influence antigen recognition. Different computational approaches, from traditional template-based modelling to modern deep-learning methods, can be used to predict antibody structures.

In this exercise, you will model an antibody from a patient with chronic lymphocytic leukemia (CLL), explore its structure using PyMOL, and investigate how sequence variation and structural features may influence antigen binding.

== Learning objectives ==

By the end of this exercise, you should be able to:

* Identify the framework and CDR regions of an antibody.
* Understand the basic principles of template-based antibody modelling.
* Generate and visualize a three-dimensional antibody model using computational tools.
* Explore how CDR sequence, structure, and electrostatic properties may influence antigen recognition.
* Identify somatic mutations and discuss their potential effects on antibody binding.
* Compare antibody structures predicted using traditional template-based and modern deep-learning approaches.

== Antibody sequences ==

We will use the variable regions of the SI135 antibody.

'''Light chain (VL):'''

<pre>
>SI135_VL
SYVVTQPPSVSVAPGKTARITCGGNNIGSKSVHWYQQKPGQAPVLVIYYDNDRPSGIPERFSGSNSGNTATLTISRVEAGDEADYYCQVWDSGSDHPWVFGGGTKLTVL
</pre>

'''Heavy chain (VH):'''

<pre>
>SI135_VH
EVQLVESGGGLVRPGGSLRLSCAASGFTFSTYSMNWVRQAPGKGLEWVSSISSSSNSIYCADSVKGRFTISRDNANNSLYMQMNSLRAEDTAVYYCARDVNTMDVWGQGILVTVLL
</pre>

== Part 1 — Identify the antibody regions ==

Go to the [https://services.healthtech.dtu.dk/services/LYRA-1.0/ Lyra] web page.

''Lyra is a tool for predicting antibody and TCR structures. It is possible to use Lyra in automated or expert mode.''

Insert the light and heavy chains in the proper text area and submit. After a few seconds you will arrive to the output page, where the alignment and all the templates are displayed.

Note that for each Region of each chain of the antibody a Template has been found.

{| class="wikitable"
! Region
! Templates
|-
| Framework
| ...
|-
| Loop 1 (CDR1)
| ...
|-
| Loop 2 (CDR2)
| ...
|-
| Loop 3 (CDR3)
| ...
|}

Note that after the Regions there is an extra template used for packing the whole structure.

Explore the output page and the different tabs - check the PDB file, Log, and manual selection ones to find this info.

Now answer the following questions:

'''Q1: what is the sequence similarity of the best H and L templates for the framework (FW)?'''

'''Q2: What is the light chain type? Is it the most common in humans?'''

=== Sequence alignment and CDR modelling ===

Focus on the sequence alignment used to generate the model. You can find it in one of the tabs in Lyra website, as well as in the PDB file itself.

'''Q3: where are the differences between target and template located? Which are the most drastic that could affect the modeling process (e.g. G->W)?'''

''hint: if you don't remeber the amino acids, you can use the Blosum62 matrix to estimate the expected effect of a substitution: the lower the number, the larger the expected impact. For sake of time, you only need to give a qualitative answer.''

'''Q4: Do the loops/CDR have the same length in target and template? What does it mean, in terms of expected accuracy?'''

=== Selection of structural templates ===

Now, look at the templates automatically selected by Lyra to build the final model.

'''Q5: How many templates were used? Which is the region for which the template is more dissimilar to the target sequence?'''

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Visualizing the antigen-binding site (ABS) ===

Download the pdb file and open it with pymol. There is a small "S" bottom right, it shows/hides the sequence. Show the whole protein as cartoon and color by chain. Select and color in red all the CDRs. (the selected residues appear as the "sele" item in the menu). The CDR definition can be seen in the alignment generated by Lyra.

'''Q6: What is the general shape of the ABS: flat, a pocket, a mountain or a groove?''' ''It can help to show the antibody as surface''

=== CDR loop structure ===

'''Q7: Which loop is the longest and protruding the most? Is this the most common case?'''

== Part 3 — Somatic mutations ==

=== Identifying mutations in the antibody sequence ===

Use the [http://www.imgt.org/3Dstructure-DB/cgi/DomainGapAlign.cgi Domain Gap Align tool] identify the somatic mutations in the V and J regions. You need to use the amino acid sequence and select human as the source organism. In PyMol, show the mutations as sticks (main chain and side chain) and color them green.

Unfortunately today the server has a planning update so the output of the tool are provided below

[[image:Imgt_GapAlign_results_VH.png|500px|center]]

[[image:Imgt_GapAlign_results_VL.png|500px|center]]

'''Q8: Which Mutations are located inside/near the CDRs? Which are the ones that are most likely to have an impact on the binding of the antibody?''' ''A qualitative answer is sufficient''

== Part 4 — Electrostatic properties of the antigen-binding site ==

=== Visualizing the electrostatic potential ===

Generate the electrostatic potentials of the Ig ('''A'''ction -> generate -> vacuum electrostatic -> protein contact potentials).

'''Q9: Identify the residues in the CDRs that mostly contribute to the charge distribution.'''

''highlight them in the sequence and look for charged residues such as K, R, D, E''

=== Modifying the electrostatic properties (optional) ===

'''Q10: (optional) Mutate some residues in order to have a positive charge in the middle of the antigen binding site, then recalculate the electrostatic potentials. Can you see the changes?'''

== Part 5 — Comparing antibody modelling approaches ==

=== LYRA versus ABodyBuilder2 ===

'''Q11: Go to the AbodyBuilder 2 site and make a new structure prediction. Download the PDB file and compare with the model generated by LYRA. Can you see some differences?'''

'''Finished!'''

File:Imgt GapAlign results VL.png

2026-10-21T03:14:47Z

Carol:

File:Imgt GapAlign results VH.png

2026-10-21T03:14:23Z

Carol:

Exercise:AntibodyStructurePrediction ans

2026-10-21T03:00:19Z

Carol:

__TOC__

= Antibody Structure Modelling — Answers =

These answers refer to the SI135 antibody and the LYRA model generated from its heavy- and light-chain variable-region sequences. Residue numbering is identified as IMGT or PDB numbering where relevant.

https://services.healthtech.dtu.dk/cgi-bin/webface2.cgi?jobid=6AD8166000282F67166ED625&wait=20

== Part 1 — Antibody modelling with LYRA ==

=== Q1. Framework template similarity ===

LYRA selected the following templates for the antibody frameworks:

{| class="wikitable"
! Chain !! Selected template !! Sequence identity !! BLOSUM score
|-
| Heavy chain (VH) || '''3H42''' || '''80.5%''' || 514
|-
| Light chain (VL) || '''1W72''' || '''85.7%''' || 518
|}

The heavy-chain candidate 2UZI has a slightly higher sequence identity (81.7%), but a lower BLOSUM score (489) than 3H42. Thus, the template with the highest percentage of identical amino acids is not necessarily the one selected for modelling. Template selection also considers sequence similarity and the compatibility of the antibody's structural features.

=== Q2. Light-chain type ===

SI135 has a '''lambda (λ) light chain'''. This is not the most common light-chain type in humans: approximately 60% of human antibodies have kappa (κ) light chains and 40% have lambda (λ) light chains, although the proportions vary.

=== Q3. Differences between the target and template sequences ===

The target and framework templates are relatively similar, but there are substitutions in both framework and CDR regions. Particularly informative differences in the selected loop-template alignments include:

{| class="wikitable"
! Region !! Template → SI135 !! Possible consequence
|-
| H1 || S → T || Conservative change between polar residues
|-
| H2 || S → N || Changes side-chain size and hydrogen-bonding possibilities
|-
| H3 || R → N || Removes a positive charge
|-
| H3 || Q → T || Changes side-chain size and hydrogen-bonding properties
|-
| L2 || D → Y || Removes a negative charge and introduces a bulky aromatic group
|-
| L3 || L → S || Replaces a hydrophobic residue with a polar residue
|-
| L3 || S → D || Introduces a negatively charged residue
|-
| L3 || A → H || Introduces a larger side chain that may be protonated
|-
| L3 || D → P || Removes a negative charge and introduces a conformationally restricted residue
|}

The '''D → Y change in L2''' is particularly non-conservative (BLOSUM62 score −3). The '''D → P change in L3''' may also matter because proline strongly restricts backbone geometry. Changes at buried framework positions can influence stability, whereas changes in exposed CDR positions can influence the chemistry and shape of the binding site. The likely effect cannot be inferred from the substitution score alone.

=== Q4. CDR lengths and modelling accuracy ===

'''All six selected LYRA loop templates have the same loop-core length as their corresponding SI135 loops.'''

{| class="wikitable"
! Loop !! PDB template !! Template sequence !! SI135 sequence !! Length
|-
| H1 || 3H42 || GFTFSSY || GFTFSTY || 7
|-
| H2 || 3H42 || SSSS || SSSN || 4
|-
| H3 || 2NR6 || VRQMD || VNTMD || 5
|-
| L1 || 2G75 || GNNIGSKS || GNNIGSKS || 8
|-
| L2 || 1W72 || DDS || YDN || 3
|-
| L3 || 3GHB || WDSGLSADW || WDSGSDHPW || 9
|}

These are the '''loop-core sequences reported in the LYRA PDB annotations'''; broader CDR definitions include additional flanking residues.

Matching lengths make template-based modelling easier because insertions or deletions are not needed within the selected loop cores. However, matching length does not guarantee matching conformation. In particular, '''L1 has template canonical structure class 7 but predicted class 8''', even though its loop-core sequence is identical. This suggests that its modelled backbone should be interpreted cautiously.

=== Q5. Number of structural templates ===

LYRA used '''five distinct PDB templates''':

{| class="wikitable"
! PDB !! Region or regions modelled
|-
| '''3H42''' || Heavy-chain framework, H1, H2 and heavy/light-chain packing
|-
| '''2NR6''' || H3
|-
| '''1W72''' || Light-chain framework and L2
|-
| '''2G75''' || L1
|-
| '''3GHB''' || L3
|}

These structures fill nine framework, loop and packing assignments, but some PDB entries are reused.

The '''lowest percentage sequence identity''' among the selected loop cores occurs in '''L2''': only 1 of its 3 residues is identical to the template (33.3%). In contrast, '''L3 has the largest absolute number of substitutions''': 4 of its 9 residues differ from the selected template (55.6% identity). Both are comparatively uncertain regions, but for different reasons.

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Q6. Shape of the antigen-binding site ===

The predicted antigen-binding surface is best described as a '''shallow pocket or groove'''. The CDR loops form an uneven surface with a depression between protruding regions rather than a completely flat plane.

The classification is qualitative: the appearance changes with the orientation of the model and how its molecular surface is displayed. The surface shape alone does not identify the antigen or establish binding specificity.

=== Q7. Longest and most protruding loop ===

'''CDR-L3 is the longest loop in the selected LYRA model''', with a 9-residue loop core (or 10 residues in the corresponding PDB loop selection including a flanking residue). CDR-L1 is the next longest in these definitions.

CDR-L3 is also a conspicuous protruding feature of the binding site. Depending on the viewing direction and whether backbone or surface protrusion is considered, '''H2 may appear to extend further than L3'''. Thus, the longest loop is not necessarily the one with the greatest height above the binding surface.

This is not the usual pattern emphasized for human antibodies, where CDR-H3 is commonly the most variable and often a prominent part of the binding site. In SI135, CDR-H3 is relatively short.

== Part 3 — Somatic mutations ==

=== Q8. Mutations within or near the CDRs ===

The germline comparison identifies the following amino acid substitutions in the CDR regions ('''IMGT numbering'''; germline residue → SI135 residue):

{| class="wikitable"
! Chain !! Region !! Mutation !! Interpretation
|-
| Heavy || CDR-H1 || S36T || Conservative polar substitution
|-
| Heavy || CDR-H2 || S63N || Changes side-chain geometry and hydrogen bonding
|-
| Heavy || CDR-H2 || '''Y64S''' || Removes a bulky aromatic side chain
|-
| Light || CDR-L2 || S65N || Changes side-chain geometry and hydrogen bonding
|-
| Light || CDR-L3 || S110G || Introduces glycine, which increases local backbone flexibility
|}

A further heavy-chain framework substitution, '''Y67C''', lies close to CDR-H2 and may influence the local structure. Other framework substitutions are present but are not necessarily directly involved in antigen contacts.

The strongest candidates for influencing antigen binding are '''Y64S in CDR-H2''' and potentially '''Y67C immediately outside this loop''', because they remove aromatic residues. The CDR substitutions S63N, S65N and S110G could also influence the local geometry or hydrogen-bonding interactions. In the BLOSUM62 matrix S → N has a positive score (+1), so it is not an especially drastic substitution by that measure alone.

Whether any mutation actually changes antigen affinity depends on its position and interactions in the three-dimensional complex. '''Mutations relative to germline should not be confused with the differences between SI135 and the LYRA structural templates.'''

== Part 4 — Electrostatic properties of the antigen-binding site ==

=== Q9. Charged CDR residues ===

The important charged side chains are lysine (K) and arginine (R), which are usually positive, and aspartate (D) and glutamate (E), which are usually negative. Histidine (H) may carry a positive charge depending on its local environment and pH.

The main contributors in the '''LYRA PDB loop selections''' are:

{| class="wikitable"
! Loop !! Charged residues !! LYRA PDB positions
|-
| H1 || No K, R, D or E || —
|-
| H2 || No K, R, D or E || —
|-
| H3 || One D (negative) || H101
|-
| L1 || One K (positive) || L31
|-
| L2 || Two D residues (negative) || L51, L53
|-
| L3 || Two D residues (negative); one potentially protonated H || L92, L95A; H at L95B
|}

The short LYRA '''L2 loop-core sequence YDN''' contains one D; the corresponding PDB loop selection includes an additional flanking D at residue L53. This explains why either one or two aspartates may be counted depending on the CDR boundaries used.

The negatively charged residues in L2 and L3 can contribute to a '''negative electrostatic patch''' around parts of the binding site. However, potential depends on solvent exposure, protonation and the three-dimensional arrangement of residues, not only on their number.

In a common electrostatic coloring convention, '''red indicates negative potential''' and '''blue indicates positive potential'''. The result from PyMOL's vacuum potential calculation is a qualitative representation.

=== Q10. Introducing a positive charge (optional) ===

Changing an exposed acidic residue near the centre of the antigen-binding site to lysine or arginine should make the local electrostatic potential more positive. For example, '''D → K''' replaces a negatively charged side chain with a positively charged one.

After recalculating the surface potential, the region around the mutation would be expected to shift towards less red or more blue, depending on the color scale used. The effect can remain local and may be less obvious when other nearby charged residues dominate the surface potential.

The reverse experiment also illustrates the same principle: replacing '''K31 in CDR-L1 with alanine''' removes a positive charge and can reduce a local positive electrostatic patch. A change in surface potential is not evidence by itself of improved or reduced antigen binding.

== Part 5 — Comparing antibody modelling approaches ==

=== Q11. LYRA versus ABodyBuilder2 ===

Both LYRA and ABodyBuilder2 predict a similar '''overall immunoglobulin fold''' for the heavy- and light-chain variable domains. The conserved framework beta-sheets are broadly consistent between the models.

Differences are most likely to be visible in the '''CDR loops''', whose local backbone conformations may vary between a template-based and a deep-learning prediction. Comparisons of SI135 models show deviations in individual loops, particularly around '''CDR-H3 and CDR-L3'''; the local conformation of other loops, including L1, may also differ.

{| class="wikitable"
! LYRA !! ABodyBuilder2
|-
| Template-based homology modelling || Deep-learning-based structure prediction
|-
| Uses selected framework and CDR structural templates || Predicts antibody structure from the amino acid sequences
|-
| Accuracy depends in part on template suitability || Reports prediction-error estimates for different structural regions
|}

When the structures are superimposed, the frameworks are expected to agree more closely than the variable loops. Any differences should be interpreted as '''model uncertainty''', not as proof that one model is correct. Neither prediction alone establishes the experimentally observed antibody conformation or its antigen specificity.

== Summary ==

* The SI135 antibody contains a lambda light chain.
* LYRA selected frameworks 3H42 (80.5% identity) and 1W72 (85.7% identity).
* Five distinct structural templates were used for framework, loop and packing modelling.
* All selected loop cores have matching target/template lengths, but CDR-L1 differs in its canonical structure classification.
* CDR-L2 has the lowest loop-core sequence identity, while CDR-L3 has the greatest number of residue substitutions.
* CDR-L3 is the longest loop; the apparent extent of protrusion depends on the representation and viewing direction.
* CDR mutations can influence binding by changing side-chain chemistry or loop conformation, but effects on affinity require experimental validation.
* The electrostatic properties of the binding site reflect the positions of charged residues, especially aspartates in the light-chain CDRs.
* Template-based and deep-learning predictions generally agree on the immunoglobulin framework but may differ in CDR conformations.

Exercise:AntibodyStructurePrediction ans

2026-10-21T02:56:35Z

Carol:

__TOC__

= Antibody Structure Modelling — Answers =

These answers refer to the SI135 antibody and the LYRA model generated from its heavy- and light-chain variable-region sequences. Residue numbering is identified as IMGT or PDB numbering where relevant.

== Part 1 — Antibody modelling with LYRA ==

=== Q1. Framework template similarity ===

LYRA selected the following templates for the antibody frameworks:

{| class="wikitable"
! Chain !! Selected template !! Sequence identity !! BLOSUM score
|-
| Heavy chain (VH) || '''3H42''' || '''80.5%''' || 514
|-
| Light chain (VL) || '''1W72''' || '''85.7%''' || 518
|}

The heavy-chain candidate 2UZI has a slightly higher sequence identity (81.7%), but a lower BLOSUM score (489) than 3H42. Thus, the template with the highest percentage of identical amino acids is not necessarily the one selected for modelling. Template selection also considers sequence similarity and the compatibility of the antibody's structural features.

=== Q2. Light-chain type ===

SI135 has a '''lambda (λ) light chain'''. This is not the most common light-chain type in humans: approximately 60% of human antibodies have kappa (κ) light chains and 40% have lambda (λ) light chains, although the proportions vary.

=== Q3. Differences between the target and template sequences ===

The target and framework templates are relatively similar, but there are substitutions in both framework and CDR regions. Particularly informative differences in the selected loop-template alignments include:

{| class="wikitable"
! Region !! Template → SI135 !! Possible consequence
|-
| H1 || S → T || Conservative change between polar residues
|-
| H2 || S → N || Changes side-chain size and hydrogen-bonding possibilities
|-
| H3 || R → N || Removes a positive charge
|-
| H3 || Q → T || Changes side-chain size and hydrogen-bonding properties
|-
| L2 || D → Y || Removes a negative charge and introduces a bulky aromatic group
|-
| L3 || L → S || Replaces a hydrophobic residue with a polar residue
|-
| L3 || S → D || Introduces a negatively charged residue
|-
| L3 || A → H || Introduces a larger side chain that may be protonated
|-
| L3 || D → P || Removes a negative charge and introduces a conformationally restricted residue
|}

The '''D → Y change in L2''' is particularly non-conservative (BLOSUM62 score −3). The '''D → P change in L3''' may also matter because proline strongly restricts backbone geometry. Changes at buried framework positions can influence stability, whereas changes in exposed CDR positions can influence the chemistry and shape of the binding site. The likely effect cannot be inferred from the substitution score alone.

=== Q4. CDR lengths and modelling accuracy ===

'''All six selected LYRA loop templates have the same loop-core length as their corresponding SI135 loops.'''

{| class="wikitable"
! Loop !! PDB template !! Template sequence !! SI135 sequence !! Length
|-
| H1 || 3H42 || GFTFSSY || GFTFSTY || 7
|-
| H2 || 3H42 || SSSS || SSSN || 4
|-
| H3 || 2NR6 || VRQMD || VNTMD || 5
|-
| L1 || 2G75 || GNNIGSKS || GNNIGSKS || 8
|-
| L2 || 1W72 || DDS || YDN || 3
|-
| L3 || 3GHB || WDSGLSADW || WDSGSDHPW || 9
|}

These are the '''loop-core sequences reported in the LYRA PDB annotations'''; broader CDR definitions include additional flanking residues.

Matching lengths make template-based modelling easier because insertions or deletions are not needed within the selected loop cores. However, matching length does not guarantee matching conformation. In particular, '''L1 has template canonical structure class 7 but predicted class 8''', even though its loop-core sequence is identical. This suggests that its modelled backbone should be interpreted cautiously.

=== Q5. Number of structural templates ===

LYRA used '''five distinct PDB templates''':

{| class="wikitable"
! PDB !! Region or regions modelled
|-
| '''3H42''' || Heavy-chain framework, H1, H2 and heavy/light-chain packing
|-
| '''2NR6''' || H3
|-
| '''1W72''' || Light-chain framework and L2
|-
| '''2G75''' || L1
|-
| '''3GHB''' || L3
|}

These structures fill nine framework, loop and packing assignments, but some PDB entries are reused.

The '''lowest percentage sequence identity''' among the selected loop cores occurs in '''L2''': only 1 of its 3 residues is identical to the template (33.3%). In contrast, '''L3 has the largest absolute number of substitutions''': 4 of its 9 residues differ from the selected template (55.6% identity). Both are comparatively uncertain regions, but for different reasons.

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Q6. Shape of the antigen-binding site ===

The predicted antigen-binding surface is best described as a '''shallow pocket or groove'''. The CDR loops form an uneven surface with a depression between protruding regions rather than a completely flat plane.

The classification is qualitative: the appearance changes with the orientation of the model and how its molecular surface is displayed. The surface shape alone does not identify the antigen or establish binding specificity.

=== Q7. Longest and most protruding loop ===

'''CDR-L3 is the longest loop in the selected LYRA model''', with a 9-residue loop core (or 10 residues in the corresponding PDB loop selection including a flanking residue). CDR-L1 is the next longest in these definitions.

CDR-L3 is also a conspicuous protruding feature of the binding site. Depending on the viewing direction and whether backbone or surface protrusion is considered, '''H2 may appear to extend further than L3'''. Thus, the longest loop is not necessarily the one with the greatest height above the binding surface.

This is not the usual pattern emphasized for human antibodies, where CDR-H3 is commonly the most variable and often a prominent part of the binding site. In SI135, CDR-H3 is relatively short.

== Part 3 — Somatic mutations ==

=== Q8. Mutations within or near the CDRs ===

The germline comparison identifies the following amino acid substitutions in the CDR regions ('''IMGT numbering'''; germline residue → SI135 residue):

{| class="wikitable"
! Chain !! Region !! Mutation !! Interpretation
|-
| Heavy || CDR-H1 || S36T || Conservative polar substitution
|-
| Heavy || CDR-H2 || S63N || Changes side-chain geometry and hydrogen bonding
|-
| Heavy || CDR-H2 || '''Y64S''' || Removes a bulky aromatic side chain
|-
| Light || CDR-L2 || S65N || Changes side-chain geometry and hydrogen bonding
|-
| Light || CDR-L3 || S110G || Introduces glycine, which increases local backbone flexibility
|}

A further heavy-chain framework substitution, '''Y67C''', lies close to CDR-H2 and may influence the local structure. Other framework substitutions are present but are not necessarily directly involved in antigen contacts.

The strongest candidates for influencing antigen binding are '''Y64S in CDR-H2''' and potentially '''Y67C immediately outside this loop''', because they remove aromatic residues. The CDR substitutions S63N, S65N and S110G could also influence the local geometry or hydrogen-bonding interactions. In the BLOSUM62 matrix S → N has a positive score (+1), so it is not an especially drastic substitution by that measure alone.

Whether any mutation actually changes antigen affinity depends on its position and interactions in the three-dimensional complex. '''Mutations relative to germline should not be confused with the differences between SI135 and the LYRA structural templates.'''

== Part 4 — Electrostatic properties of the antigen-binding site ==

=== Q9. Charged CDR residues ===

The important charged side chains are lysine (K) and arginine (R), which are usually positive, and aspartate (D) and glutamate (E), which are usually negative. Histidine (H) may carry a positive charge depending on its local environment and pH.

The main contributors in the '''LYRA PDB loop selections''' are:

{| class="wikitable"
! Loop !! Charged residues !! LYRA PDB positions
|-
| H1 || No K, R, D or E || —
|-
| H2 || No K, R, D or E || —
|-
| H3 || One D (negative) || H101
|-
| L1 || One K (positive) || L31
|-
| L2 || Two D residues (negative) || L51, L53
|-
| L3 || Two D residues (negative); one potentially protonated H || L92, L95A; H at L95B
|}

The short LYRA '''L2 loop-core sequence YDN''' contains one D; the corresponding PDB loop selection includes an additional flanking D at residue L53. This explains why either one or two aspartates may be counted depending on the CDR boundaries used.

The negatively charged residues in L2 and L3 can contribute to a '''negative electrostatic patch''' around parts of the binding site. However, potential depends on solvent exposure, protonation and the three-dimensional arrangement of residues, not only on their number.

In a common electrostatic coloring convention, '''red indicates negative potential''' and '''blue indicates positive potential'''. The result from PyMOL's vacuum potential calculation is a qualitative representation.

=== Q10. Introducing a positive charge (optional) ===

Changing an exposed acidic residue near the centre of the antigen-binding site to lysine or arginine should make the local electrostatic potential more positive. For example, '''D → K''' replaces a negatively charged side chain with a positively charged one.

After recalculating the surface potential, the region around the mutation would be expected to shift towards less red or more blue, depending on the color scale used. The effect can remain local and may be less obvious when other nearby charged residues dominate the surface potential.

The reverse experiment also illustrates the same principle: replacing '''K31 in CDR-L1 with alanine''' removes a positive charge and can reduce a local positive electrostatic patch. A change in surface potential is not evidence by itself of improved or reduced antigen binding.

== Part 5 — Comparing antibody modelling approaches ==

=== Q11. LYRA versus ABodyBuilder2 ===

Both LYRA and ABodyBuilder2 predict a similar '''overall immunoglobulin fold''' for the heavy- and light-chain variable domains. The conserved framework beta-sheets are broadly consistent between the models.

Differences are most likely to be visible in the '''CDR loops''', whose local backbone conformations may vary between a template-based and a deep-learning prediction. Comparisons of SI135 models show deviations in individual loops, particularly around '''CDR-H3 and CDR-L3'''; the local conformation of other loops, including L1, may also differ.

{| class="wikitable"
! LYRA !! ABodyBuilder2
|-
| Template-based homology modelling || Deep-learning-based structure prediction
|-
| Uses selected framework and CDR structural templates || Predicts antibody structure from the amino acid sequences
|-
| Accuracy depends in part on template suitability || Reports prediction-error estimates for different structural regions
|}

When the structures are superimposed, the frameworks are expected to agree more closely than the variable loops. Any differences should be interpreted as '''model uncertainty''', not as proof that one model is correct. Neither prediction alone establishes the experimentally observed antibody conformation or its antigen specificity.

== Summary ==

* The SI135 antibody contains a lambda light chain.
* LYRA selected frameworks 3H42 (80.5% identity) and 1W72 (85.7% identity).
* Five distinct structural templates were used for framework, loop and packing modelling.
* All selected loop cores have matching target/template lengths, but CDR-L1 differs in its canonical structure classification.
* CDR-L2 has the lowest loop-core sequence identity, while CDR-L3 has the greatest number of residue substitutions.
* CDR-L3 is the longest loop; the apparent extent of protrusion depends on the representation and viewing direction.
* CDR mutations can influence binding by changing side-chain chemistry or loop conformation, but effects on affinity require experimental validation.
* The electrostatic properties of the binding site reflect the positions of charged residues, especially aspartates in the light-chain CDRs.
* Template-based and deep-learning predictions generally agree on the immunoglobulin framework but may differ in CDR conformations.

Exercise:AntibodyStructurePrediction ans

2026-10-21T02:33:18Z

Carol: /* Q6. Shape of the antigen-binding site */

__TOC__

= Antibody Structure Modelling — Answers =

''SI135 CLL antibody modelling exercise.''

== Part 1 — Antibody modelling with LYRA ==

=== Q1. Framework template selection ===

'''Answer:'''

LYRA selected the following framework templates:

{| class="wikitable"
! Chain
! Selected template
! Sequence identity
! BLOSUM score
|-
| Heavy (VH)
| '''3H42'''
| '''80.5%'''
| 514
|-
| Lambda light (VL)
| '''1W72'''
| '''85.7%'''
| 518
|}

Interestingly, 2UZI has a slightly higher sequence identity to the heavy chain (81.7%), but LYRA selects 3H42, which has a higher BLOSUM score and compatible canonical structures.

This illustrates that template selection depends on more than sequence identity alone.

=== Q2. Light-chain type ===

'''Answer:'''

The SI135 antibody has a '''lambda (λ) light chain'''.

Lambda is less common than kappa (κ) in human antibodies. Approximately 60% of human antibodies have kappa light chains and 40% have lambda light chains.

=== Q3. Sequence differences between target and templates ===

'''Answer:'''

The target antibody is relatively similar to the selected framework templates, but several substitutions are observed in both framework and CDR regions.

Some examples from the LYRA alignments include:

{| class="wikitable"
! Region
! Template → Target
! Interpretation
|-
| Heavy-chain framework
| K → R
| Conservative substitution between positively charged residues
|-
| CDR-H1
| S → T
| Conservative polar substitution
|-
| CDR-H2
| S → N
| Changes side-chain size and hydrogen-bonding possibilities
|-
| CDR-H3
| R → N
| Removes a positive charge
|-
| CDR-H3
| Q → T
| Changes side-chain size and hydrogen-bonding properties
|-
| CDR-L2
| D → Y
| Removes a negative charge and introduces a bulky aromatic residue
|-
| CDR-L3
| L → S
| Changes from hydrophobic to polar
|-
| CDR-L3
| S → D
| Introduces a negative charge
|-
| CDR-L3
| A → H
| Introduces a larger, potentially ionizable side chain
|-
| CDR-L3
| D → P
| Removes a negative charge and introduces a residue that restricts backbone conformation
|}

The CDR-L2 substitution D → Y and the CDR-L3 substitution D → P are particularly interesting because they involve substantial changes in physicochemical properties.

In the light-chain framework alignment, a D → K substitution near the C-terminus also reverses the side-chain charge.

'''Note:''' The impact of a substitution depends not only on its BLOSUM score but also on its structural location and whether the residue is buried or exposed.

=== Q4. CDR loop lengths and expected accuracy ===

'''Answer:'''

All six selected CDR templates have the same loop length as the corresponding SI135 target loops.

{| class="wikitable"
! Loop
! Template
! Template sequence
! Target sequence
! Length
|-
| H1
| 3H42
| GFTFSSY
| GFTFSTY
| 7
|-
| H2
| 3H42
| SSSS
| SSSN
| 4
|-
| H3
| 2NR6
| VRQMD
| VNTMD
| 5
|-
| L1
| 2G75
| GNNIGSKS
| GNNIGSKS
| 8
|-
| L2
| 1W72
| DDS
| YDN
| 3
|-
| L3
| 3GHB
| WDSGLSADW
| WDSGSDHPW
| 9
|}

''These are the loop-core lengths reported by LYRA. Other CDR definitions may include additional residues.''

Matching loop lengths generally make template-based modelling easier because no insertions or deletions are required in the loop core.

However, there is one interesting exception in the canonical structure classification.

For CDR-L1:

* '''Template canonical structure: 7'''
* '''Predicted canonical structure: 8'''

Thus, although L1 has the same sequence length and an identical loop-core sequence in the selected template, the predicted canonical structure class is different.

This introduces additional uncertainty in the loop conformation.

'''Note:''' Matching loop length does not guarantee that the target and template have the same three-dimensional conformation.

=== Q5. Number of templates and most dissimilar region ===

'''Answer:'''

LYRA uses '''five distinct PDB templates''' to construct the SI135 antibody model:

{| class="wikitable"
! PDB template
! Regions modelled
|-
| '''3H42'''
| Heavy-chain framework, H1, H2 and heavy/light chain packing
|-
| '''2NR6'''
| H3
|-
| '''1W72'''
| Light-chain framework and L2
|-
| '''2G75'''
| L1
|-
| '''3GHB'''
| L3
|}

A total of eight framework/loop regions are modelled, but several use the same structural template.

The sequence identity within each selected loop core is:

{| class="wikitable"
! Loop
! Identical residues
! Identity
|-
| H1
| 6/7
| 85.7%
|-
| H2
| 3/4
| 75.0%
|-
| H3
| 3/5
| 60.0%
|-
| L1
| 8/8
| 100%
|-
| L2
| 1/3
| '''33.3%'''
|-
| L3
| 5/9
| 55.6%
|}

'''CDR-L2 has the lowest direct sequence identity''' between its selected loop template and the target.

CDR-L3 also contains several substitutions, including the potentially important D → P change.

The loop identities above are calculated directly from the sequences reported in the PDB annotations, not from the overall framework-template scores.

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Q6. Shape of the antigen-binding site ===

'''Answer:'''

The antigen-binding site is formed by the six CDR loops.

Its overall shape can be classified by displaying the predicted antibody as a molecular surface and examining the arrangement of these loops.

'''SI135 classification:''' Pocket-like

Note that this classification is qualitative.

=== Q7. Longest and most protruding CDR ===

'''Answer:'''

According to LYRA, the six loop-core lengths are:

{| class="wikitable"
! Loop
! Length
|-
| H1
| 7
|-
| H2
| 4
|-
| H3
| 5
|-
| L1
| 8
|-
| L2
| 3
|-
| '''L3'''
| '''9'''
|}

The '''longest loop is CDR-L3'''.

This is notable because CDR-H3 is often the longest and most structurally variable CDR in human antibodies.

The loop that protrudes furthest should be determined from the three-dimensional structure, since length alone does not determine protrusion.

'''Most protruding loop:''' To be confirmed by visual inspection in PyMOL.

== Part 3 — Somatic mutations ==

=== Q8. Somatic mutations near the CDRs ===

'''Expected answer:'''

Somatic mutations must be identified by comparing the SI135 sequence with the closest germline immunoglobulin genes using IMGT/DomainGapAlign.

The target-template substitutions identified by LYRA are not necessarily somatic mutations, because the structural templates are unrelated antibody sequences rather than the germline ancestors of SI135.

Mutations located inside the CDRs or affecting the residues that support them are particularly interesting.

The most likely candidates to affect binding are substitutions that:

* Change the charge of an exposed residue.
* Introduce or remove aromatic residues.
* Affect hydrogen bonding.
* Change loop geometry or flexibility.

'''SI135-specific mutation list:''' Requires the IMGT germline alignment.

== Part 4 — Electrostatic properties ==

=== Q9. Charged CDR residues ===

'''Answer:'''

The LYRA loop definitions allow us to identify the following charged or potentially charged residues.

{| class="wikitable"
! Loop
! Sequence
! Relevant residues
|-
| H1
| GFTFSTY
| None of K, R, D or E
|-
| H2
| SSSN
| None of K, R, D or E
|-
| H3
| VNTMD
| D (negative)
|-
| L1
| GNNIGSKS
| K (positive)
|-
| L2
| YDN
| D (negative)
|-
| L3
| WDSGSDHPW
| Two D residues (negative); H may be protonated
|}

The principal charged residues within these CDR definitions are therefore:

* One aspartate in H3.
* One lysine in L1.
* One aspartate in L2.
* Two aspartates in L3.

The light-chain loops contain several negatively charged residues, which may contribute to negatively charged patches on the antigen-binding surface.

However, the actual electrostatic distribution depends on the spatial arrangement of these residues.

'''Residue numbering in the provided LYRA PDB:'''

{| class="wikitable"
! Loop
! Residue
! PDB position
|-
| H3
| Aspartate
| H101
|-
| L1
| Lysine
| L31
|-
| L2
| Aspartate
| L51
|-
| L3
| Aspartate
| L92
|-
| L3
| Aspartate
| L95A
|-
| L3
| Histidine
| L95B
|}

You can highlight the main charged CDR residues using the following PyMOL commands:

<pre>
select charged_cdrs, (chain H and resi 101) or (chain L and resi 31+51+92+95A)
show sticks, charged_cdrs
color yellow, charged_cdrs
</pre>

''These residue numbers refer specifically to the supplied LYRA PDB file.''

=== Q10. Introducing a positive charge (optional) ===

'''Expected answer:'''

Replacing an exposed negatively charged residue with lysine or arginine should make its local electrostatic environment more positive.

For example, a D → K mutation at a solvent-exposed CDR position would reverse the formal side-chain charge.

After recalculating the electrostatic potential, the surface coloring near the mutation should change.

'''Actual changes:''' Require performing the mutation and recalculating the potential in PyMOL.

A change in electrostatic potential does not necessarily improve antigen binding.

== Part 5 — Comparing antibody modelling approaches ==

=== Q11. LYRA versus ABodyBuilder2 ===

'''Expected answer:'''

LYRA and ABodyBuilder2 use different approaches to structure prediction.

{| class="wikitable"
! LYRA
! ABodyBuilder2
|-
| Template-based homology modelling
| Deep-learning-based prediction
|-
| Uses experimentally determined structural templates
| Uses a trained neural network
|-
| Models framework and loops using selected templates
| Predicts the antibody structure from sequence
|-
| Depends on availability and suitability of structural templates
| Provides prediction-error estimates
|}

Both methods should produce a similar overall immunoglobulin fold.

Differences are expected to be more pronounced in variable loops, particularly where the sequence similarity to the LYRA template is low or the canonical structures differ.

For SI135, potentially interesting regions include:

* CDR-L1, because of the canonical structure mismatch.
* CDR-L2, because of its low sequence identity with the selected loop template.
* CDR-L3, because it contains several non-conservative substitutions.
* CDR-H3, because antibody H3 loops can have diverse conformations.

'''Actual differences between the two SI135 predictions:''' Require the ABodyBuilder2 model for structural superposition.

In PyMOL, the models can be superimposed using:

<pre>
align ABodyBuilder2, LYRA
</pre>

The object names should be adapted to the names used in the PyMOL session.

== Take-home messages ==

* The SI135 antibody has a lambda light chain.
* The selected framework templates show relatively high sequence identity: 80.5% for VH and 85.7% for VL.
* Five distinct structural templates were used to build the antibody model.
* All six loop templates have the same lengths as their corresponding target loops.
* CDR-L1 has a mismatch in canonical structure classification despite matching loop length.
* CDR-L2 has the lowest sequence identity to its selected loop template.
* CDR-L3 is the longest loop in the LYRA model.
* Several negatively charged residues are present in the CDRs, especially in the light chain.
* Template similarity does not guarantee structural accuracy, particularly in flexible loops.

== References ==

* [https://services.healthtech.dtu.dk/services/LYRA-1.0/ LYRA]
* [https://www.imgt.org/ IMGT]
* [https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/abodybuilder2/ ABodyBuilder2]

Exercise:AntibodyStructurePrediction ans

2026-10-21T02:32:21Z

Carol: /* Q6. Shape of the antigen-binding site */

__TOC__

= Antibody Structure Modelling — Answers =

''SI135 CLL antibody modelling exercise.''

== Part 1 — Antibody modelling with LYRA ==

=== Q1. Framework template selection ===

'''Answer:'''

LYRA selected the following framework templates:

{| class="wikitable"
! Chain
! Selected template
! Sequence identity
! BLOSUM score
|-
| Heavy (VH)
| '''3H42'''
| '''80.5%'''
| 514
|-
| Lambda light (VL)
| '''1W72'''
| '''85.7%'''
| 518
|}

Interestingly, 2UZI has a slightly higher sequence identity to the heavy chain (81.7%), but LYRA selects 3H42, which has a higher BLOSUM score and compatible canonical structures.

This illustrates that template selection depends on more than sequence identity alone.

=== Q2. Light-chain type ===

'''Answer:'''

The SI135 antibody has a '''lambda (λ) light chain'''.

Lambda is less common than kappa (κ) in human antibodies. Approximately 60% of human antibodies have kappa light chains and 40% have lambda light chains.

=== Q3. Sequence differences between target and templates ===

'''Answer:'''

The target antibody is relatively similar to the selected framework templates, but several substitutions are observed in both framework and CDR regions.

Some examples from the LYRA alignments include:

{| class="wikitable"
! Region
! Template → Target
! Interpretation
|-
| Heavy-chain framework
| K → R
| Conservative substitution between positively charged residues
|-
| CDR-H1
| S → T
| Conservative polar substitution
|-
| CDR-H2
| S → N
| Changes side-chain size and hydrogen-bonding possibilities
|-
| CDR-H3
| R → N
| Removes a positive charge
|-
| CDR-H3
| Q → T
| Changes side-chain size and hydrogen-bonding properties
|-
| CDR-L2
| D → Y
| Removes a negative charge and introduces a bulky aromatic residue
|-
| CDR-L3
| L → S
| Changes from hydrophobic to polar
|-
| CDR-L3
| S → D
| Introduces a negative charge
|-
| CDR-L3
| A → H
| Introduces a larger, potentially ionizable side chain
|-
| CDR-L3
| D → P
| Removes a negative charge and introduces a residue that restricts backbone conformation
|}

The CDR-L2 substitution D → Y and the CDR-L3 substitution D → P are particularly interesting because they involve substantial changes in physicochemical properties.

In the light-chain framework alignment, a D → K substitution near the C-terminus also reverses the side-chain charge.

'''Note:''' The impact of a substitution depends not only on its BLOSUM score but also on its structural location and whether the residue is buried or exposed.

=== Q4. CDR loop lengths and expected accuracy ===

'''Answer:'''

All six selected CDR templates have the same loop length as the corresponding SI135 target loops.

{| class="wikitable"
! Loop
! Template
! Template sequence
! Target sequence
! Length
|-
| H1
| 3H42
| GFTFSSY
| GFTFSTY
| 7
|-
| H2
| 3H42
| SSSS
| SSSN
| 4
|-
| H3
| 2NR6
| VRQMD
| VNTMD
| 5
|-
| L1
| 2G75
| GNNIGSKS
| GNNIGSKS
| 8
|-
| L2
| 1W72
| DDS
| YDN
| 3
|-
| L3
| 3GHB
| WDSGLSADW
| WDSGSDHPW
| 9
|}

''These are the loop-core lengths reported by LYRA. Other CDR definitions may include additional residues.''

Matching loop lengths generally make template-based modelling easier because no insertions or deletions are required in the loop core.

However, there is one interesting exception in the canonical structure classification.

For CDR-L1:

* '''Template canonical structure: 7'''
* '''Predicted canonical structure: 8'''

Thus, although L1 has the same sequence length and an identical loop-core sequence in the selected template, the predicted canonical structure class is different.

This introduces additional uncertainty in the loop conformation.

'''Note:''' Matching loop length does not guarantee that the target and template have the same three-dimensional conformation.

=== Q5. Number of templates and most dissimilar region ===

'''Answer:'''

LYRA uses '''five distinct PDB templates''' to construct the SI135 antibody model:

{| class="wikitable"
! PDB template
! Regions modelled
|-
| '''3H42'''
| Heavy-chain framework, H1, H2 and heavy/light chain packing
|-
| '''2NR6'''
| H3
|-
| '''1W72'''
| Light-chain framework and L2
|-
| '''2G75'''
| L1
|-
| '''3GHB'''
| L3
|}

A total of eight framework/loop regions are modelled, but several use the same structural template.

The sequence identity within each selected loop core is:

{| class="wikitable"
! Loop
! Identical residues
! Identity
|-
| H1
| 6/7
| 85.7%
|-
| H2
| 3/4
| 75.0%
|-
| H3
| 3/5
| 60.0%
|-
| L1
| 8/8
| 100%
|-
| L2
| 1/3
| '''33.3%'''
|-
| L3
| 5/9
| 55.6%
|}

'''CDR-L2 has the lowest direct sequence identity''' between its selected loop template and the target.

CDR-L3 also contains several substitutions, including the potentially important D → P change.

The loop identities above are calculated directly from the sequences reported in the PDB annotations, not from the overall framework-template scores.

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Q6. Shape of the antigen-binding site ===

'''Expected answer:'''

The antigen-binding site is formed by the six CDR loops.

Its overall shape should be classified by displaying the predicted antibody as a molecular surface and examining the arrangement of these loops.

'''SI135 classification:''' Pocket-like

Note that this classification is qualitative.

=== Q7. Longest and most protruding CDR ===

'''Answer:'''

According to LYRA, the six loop-core lengths are:

{| class="wikitable"
! Loop
! Length
|-
| H1
| 7
|-
| H2
| 4
|-
| H3
| 5
|-
| L1
| 8
|-
| L2
| 3
|-
| '''L3'''
| '''9'''
|}

The '''longest loop is CDR-L3'''.

This is notable because CDR-H3 is often the longest and most structurally variable CDR in human antibodies.

The loop that protrudes furthest should be determined from the three-dimensional structure, since length alone does not determine protrusion.

'''Most protruding loop:''' To be confirmed by visual inspection in PyMOL.

== Part 3 — Somatic mutations ==

=== Q8. Somatic mutations near the CDRs ===

'''Expected answer:'''

Somatic mutations must be identified by comparing the SI135 sequence with the closest germline immunoglobulin genes using IMGT/DomainGapAlign.

The target-template substitutions identified by LYRA are not necessarily somatic mutations, because the structural templates are unrelated antibody sequences rather than the germline ancestors of SI135.

Mutations located inside the CDRs or affecting the residues that support them are particularly interesting.

The most likely candidates to affect binding are substitutions that:

* Change the charge of an exposed residue.
* Introduce or remove aromatic residues.
* Affect hydrogen bonding.
* Change loop geometry or flexibility.

'''SI135-specific mutation list:''' Requires the IMGT germline alignment.

== Part 4 — Electrostatic properties ==

=== Q9. Charged CDR residues ===

'''Answer:'''

The LYRA loop definitions allow us to identify the following charged or potentially charged residues.

{| class="wikitable"
! Loop
! Sequence
! Relevant residues
|-
| H1
| GFTFSTY
| None of K, R, D or E
|-
| H2
| SSSN
| None of K, R, D or E
|-
| H3
| VNTMD
| D (negative)
|-
| L1
| GNNIGSKS
| K (positive)
|-
| L2
| YDN
| D (negative)
|-
| L3
| WDSGSDHPW
| Two D residues (negative); H may be protonated
|}

The principal charged residues within these CDR definitions are therefore:

* One aspartate in H3.
* One lysine in L1.
* One aspartate in L2.
* Two aspartates in L3.

The light-chain loops contain several negatively charged residues, which may contribute to negatively charged patches on the antigen-binding surface.

However, the actual electrostatic distribution depends on the spatial arrangement of these residues.

'''Residue numbering in the provided LYRA PDB:'''

{| class="wikitable"
! Loop
! Residue
! PDB position
|-
| H3
| Aspartate
| H101
|-
| L1
| Lysine
| L31
|-
| L2
| Aspartate
| L51
|-
| L3
| Aspartate
| L92
|-
| L3
| Aspartate
| L95A
|-
| L3
| Histidine
| L95B
|}

You can highlight the main charged CDR residues using the following PyMOL commands:

<pre>
select charged_cdrs, (chain H and resi 101) or (chain L and resi 31+51+92+95A)
show sticks, charged_cdrs
color yellow, charged_cdrs
</pre>

''These residue numbers refer specifically to the supplied LYRA PDB file.''

=== Q10. Introducing a positive charge (optional) ===

'''Expected answer:'''

Replacing an exposed negatively charged residue with lysine or arginine should make its local electrostatic environment more positive.

For example, a D → K mutation at a solvent-exposed CDR position would reverse the formal side-chain charge.

After recalculating the electrostatic potential, the surface coloring near the mutation should change.

'''Actual changes:''' Require performing the mutation and recalculating the potential in PyMOL.

A change in electrostatic potential does not necessarily improve antigen binding.

== Part 5 — Comparing antibody modelling approaches ==

=== Q11. LYRA versus ABodyBuilder2 ===

'''Expected answer:'''

LYRA and ABodyBuilder2 use different approaches to structure prediction.

{| class="wikitable"
! LYRA
! ABodyBuilder2
|-
| Template-based homology modelling
| Deep-learning-based prediction
|-
| Uses experimentally determined structural templates
| Uses a trained neural network
|-
| Models framework and loops using selected templates
| Predicts the antibody structure from sequence
|-
| Depends on availability and suitability of structural templates
| Provides prediction-error estimates
|}

Both methods should produce a similar overall immunoglobulin fold.

Differences are expected to be more pronounced in variable loops, particularly where the sequence similarity to the LYRA template is low or the canonical structures differ.

For SI135, potentially interesting regions include:

* CDR-L1, because of the canonical structure mismatch.
* CDR-L2, because of its low sequence identity with the selected loop template.
* CDR-L3, because it contains several non-conservative substitutions.
* CDR-H3, because antibody H3 loops can have diverse conformations.

'''Actual differences between the two SI135 predictions:''' Require the ABodyBuilder2 model for structural superposition.

In PyMOL, the models can be superimposed using:

<pre>
align ABodyBuilder2, LYRA
</pre>

The object names should be adapted to the names used in the PyMOL session.

== Take-home messages ==

* The SI135 antibody has a lambda light chain.
* The selected framework templates show relatively high sequence identity: 80.5% for VH and 85.7% for VL.
* Five distinct structural templates were used to build the antibody model.
* All six loop templates have the same lengths as their corresponding target loops.
* CDR-L1 has a mismatch in canonical structure classification despite matching loop length.
* CDR-L2 has the lowest sequence identity to its selected loop template.
* CDR-L3 is the longest loop in the LYRA model.
* Several negatively charged residues are present in the CDRs, especially in the light chain.
* Template similarity does not guarantee structural accuracy, particularly in flexible loops.

== References ==

* [https://services.healthtech.dtu.dk/services/LYRA-1.0/ LYRA]
* [https://www.imgt.org/ IMGT]
* [https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/abodybuilder2/ ABodyBuilder2]

Exercise:AntibodyStructurePrediction ans

2026-10-21T02:30:59Z

Carol: /* Q9. Charged CDR residues */

__TOC__

= Antibody Structure Modelling — Answers =

''SI135 CLL antibody modelling exercise.''

== Part 1 — Antibody modelling with LYRA ==

=== Q1. Framework template selection ===

'''Answer:'''

LYRA selected the following framework templates:

{| class="wikitable"
! Chain
! Selected template
! Sequence identity
! BLOSUM score
|-
| Heavy (VH)
| '''3H42'''
| '''80.5%'''
| 514
|-
| Lambda light (VL)
| '''1W72'''
| '''85.7%'''
| 518
|}

Interestingly, 2UZI has a slightly higher sequence identity to the heavy chain (81.7%), but LYRA selects 3H42, which has a higher BLOSUM score and compatible canonical structures.

This illustrates that template selection depends on more than sequence identity alone.

=== Q2. Light-chain type ===

'''Answer:'''

The SI135 antibody has a '''lambda (λ) light chain'''.

Lambda is less common than kappa (κ) in human antibodies. Approximately 60% of human antibodies have kappa light chains and 40% have lambda light chains.

=== Q3. Sequence differences between target and templates ===

'''Answer:'''

The target antibody is relatively similar to the selected framework templates, but several substitutions are observed in both framework and CDR regions.

Some examples from the LYRA alignments include:

{| class="wikitable"
! Region
! Template → Target
! Interpretation
|-
| Heavy-chain framework
| K → R
| Conservative substitution between positively charged residues
|-
| CDR-H1
| S → T
| Conservative polar substitution
|-
| CDR-H2
| S → N
| Changes side-chain size and hydrogen-bonding possibilities
|-
| CDR-H3
| R → N
| Removes a positive charge
|-
| CDR-H3
| Q → T
| Changes side-chain size and hydrogen-bonding properties
|-
| CDR-L2
| D → Y
| Removes a negative charge and introduces a bulky aromatic residue
|-
| CDR-L3
| L → S
| Changes from hydrophobic to polar
|-
| CDR-L3
| S → D
| Introduces a negative charge
|-
| CDR-L3
| A → H
| Introduces a larger, potentially ionizable side chain
|-
| CDR-L3
| D → P
| Removes a negative charge and introduces a residue that restricts backbone conformation
|}

The CDR-L2 substitution D → Y and the CDR-L3 substitution D → P are particularly interesting because they involve substantial changes in physicochemical properties.

In the light-chain framework alignment, a D → K substitution near the C-terminus also reverses the side-chain charge.

'''Note:''' The impact of a substitution depends not only on its BLOSUM score but also on its structural location and whether the residue is buried or exposed.

=== Q4. CDR loop lengths and expected accuracy ===

'''Answer:'''

All six selected CDR templates have the same loop length as the corresponding SI135 target loops.

{| class="wikitable"
! Loop
! Template
! Template sequence
! Target sequence
! Length
|-
| H1
| 3H42
| GFTFSSY
| GFTFSTY
| 7
|-
| H2
| 3H42
| SSSS
| SSSN
| 4
|-
| H3
| 2NR6
| VRQMD
| VNTMD
| 5
|-
| L1
| 2G75
| GNNIGSKS
| GNNIGSKS
| 8
|-
| L2
| 1W72
| DDS
| YDN
| 3
|-
| L3
| 3GHB
| WDSGLSADW
| WDSGSDHPW
| 9
|}

''These are the loop-core lengths reported by LYRA. Other CDR definitions may include additional residues.''

Matching loop lengths generally make template-based modelling easier because no insertions or deletions are required in the loop core.

However, there is one interesting exception in the canonical structure classification.

For CDR-L1:

* '''Template canonical structure: 7'''
* '''Predicted canonical structure: 8'''

Thus, although L1 has the same sequence length and an identical loop-core sequence in the selected template, the predicted canonical structure class is different.

This introduces additional uncertainty in the loop conformation.

'''Note:''' Matching loop length does not guarantee that the target and template have the same three-dimensional conformation.

=== Q5. Number of templates and most dissimilar region ===

'''Answer:'''

LYRA uses '''five distinct PDB templates''' to construct the SI135 antibody model:

{| class="wikitable"
! PDB template
! Regions modelled
|-
| '''3H42'''
| Heavy-chain framework, H1, H2 and heavy/light chain packing
|-
| '''2NR6'''
| H3
|-
| '''1W72'''
| Light-chain framework and L2
|-
| '''2G75'''
| L1
|-
| '''3GHB'''
| L3
|}

A total of eight framework/loop regions are modelled, but several use the same structural template.

The sequence identity within each selected loop core is:

{| class="wikitable"
! Loop
! Identical residues
! Identity
|-
| H1
| 6/7
| 85.7%
|-
| H2
| 3/4
| 75.0%
|-
| H3
| 3/5
| 60.0%
|-
| L1
| 8/8
| 100%
|-
| L2
| 1/3
| '''33.3%'''
|-
| L3
| 5/9
| 55.6%
|}

'''CDR-L2 has the lowest direct sequence identity''' between its selected loop template and the target.

CDR-L3 also contains several substitutions, including the potentially important D → P change.

The loop identities above are calculated directly from the sequences reported in the PDB annotations, not from the overall framework-template scores.

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Q6. Shape of the antigen-binding site ===

'''Expected answer:'''

The antigen-binding site is formed by the six CDR loops.

Its overall shape should be classified by displaying the predicted antibody as a molecular surface and examining the arrangement of these loops.

'''SI135 classification:''' To be confirmed from the PyMOL surface representation.

The LYRA sequence alignments do not establish whether the binding site is flat, pocket-like, groove-like or mountain-like.

The classification is qualitative, and students should justify their answer based on the model.

=== Q7. Longest and most protruding CDR ===

'''Answer:'''

According to LYRA, the six loop-core lengths are:

{| class="wikitable"
! Loop
! Length
|-
| H1
| 7
|-
| H2
| 4
|-
| H3
| 5
|-
| L1
| 8
|-
| L2
| 3
|-
| '''L3'''
| '''9'''
|}

The '''longest loop is CDR-L3'''.

This is notable because CDR-H3 is often the longest and most structurally variable CDR in human antibodies.

The loop that protrudes furthest should be determined from the three-dimensional structure, since length alone does not determine protrusion.

'''Most protruding loop:''' To be confirmed by visual inspection in PyMOL.

== Part 3 — Somatic mutations ==

=== Q8. Somatic mutations near the CDRs ===

'''Expected answer:'''

Somatic mutations must be identified by comparing the SI135 sequence with the closest germline immunoglobulin genes using IMGT/DomainGapAlign.

The target-template substitutions identified by LYRA are not necessarily somatic mutations, because the structural templates are unrelated antibody sequences rather than the germline ancestors of SI135.

Mutations located inside the CDRs or affecting the residues that support them are particularly interesting.

The most likely candidates to affect binding are substitutions that:

* Change the charge of an exposed residue.
* Introduce or remove aromatic residues.
* Affect hydrogen bonding.
* Change loop geometry or flexibility.

'''SI135-specific mutation list:''' Requires the IMGT germline alignment.

== Part 4 — Electrostatic properties ==

=== Q9. Charged CDR residues ===

'''Answer:'''

The LYRA loop definitions allow us to identify the following charged or potentially charged residues.

{| class="wikitable"
! Loop
! Sequence
! Relevant residues
|-
| H1
| GFTFSTY
| None of K, R, D or E
|-
| H2
| SSSN
| None of K, R, D or E
|-
| H3
| VNTMD
| D (negative)
|-
| L1
| GNNIGSKS
| K (positive)
|-
| L2
| YDN
| D (negative)
|-
| L3
| WDSGSDHPW
| Two D residues (negative); H may be protonated
|}

The principal charged residues within these CDR definitions are therefore:

* One aspartate in H3.
* One lysine in L1.
* One aspartate in L2.
* Two aspartates in L3.

The light-chain loops contain several negatively charged residues, which may contribute to negatively charged patches on the antigen-binding surface.

However, the actual electrostatic distribution depends on the spatial arrangement of these residues.

'''Residue numbering in the provided LYRA PDB:'''

{| class="wikitable"
! Loop
! Residue
! PDB position
|-
| H3
| Aspartate
| H101
|-
| L1
| Lysine
| L31
|-
| L2
| Aspartate
| L51
|-
| L3
| Aspartate
| L92
|-
| L3
| Aspartate
| L95A
|-
| L3
| Histidine
| L95B
|}

You can highlight the main charged CDR residues using the following PyMOL commands:

<pre>
select charged_cdrs, (chain H and resi 101) or (chain L and resi 31+51+92+95A)
show sticks, charged_cdrs
color yellow, charged_cdrs
</pre>

''These residue numbers refer specifically to the supplied LYRA PDB file.''

=== Q10. Introducing a positive charge (optional) ===

'''Expected answer:'''

Replacing an exposed negatively charged residue with lysine or arginine should make its local electrostatic environment more positive.

For example, a D → K mutation at a solvent-exposed CDR position would reverse the formal side-chain charge.

After recalculating the electrostatic potential, the surface coloring near the mutation should change.

'''Actual changes:''' Require performing the mutation and recalculating the potential in PyMOL.

A change in electrostatic potential does not necessarily improve antigen binding.

== Part 5 — Comparing antibody modelling approaches ==

=== Q11. LYRA versus ABodyBuilder2 ===

'''Expected answer:'''

LYRA and ABodyBuilder2 use different approaches to structure prediction.

{| class="wikitable"
! LYRA
! ABodyBuilder2
|-
| Template-based homology modelling
| Deep-learning-based prediction
|-
| Uses experimentally determined structural templates
| Uses a trained neural network
|-
| Models framework and loops using selected templates
| Predicts the antibody structure from sequence
|-
| Depends on availability and suitability of structural templates
| Provides prediction-error estimates
|}

Both methods should produce a similar overall immunoglobulin fold.

Differences are expected to be more pronounced in variable loops, particularly where the sequence similarity to the LYRA template is low or the canonical structures differ.

For SI135, potentially interesting regions include:

* CDR-L1, because of the canonical structure mismatch.
* CDR-L2, because of its low sequence identity with the selected loop template.
* CDR-L3, because it contains several non-conservative substitutions.
* CDR-H3, because antibody H3 loops can have diverse conformations.

'''Actual differences between the two SI135 predictions:''' Require the ABodyBuilder2 model for structural superposition.

In PyMOL, the models can be superimposed using:

<pre>
align ABodyBuilder2, LYRA
</pre>

The object names should be adapted to the names used in the PyMOL session.

== Take-home messages ==

* The SI135 antibody has a lambda light chain.
* The selected framework templates show relatively high sequence identity: 80.5% for VH and 85.7% for VL.
* Five distinct structural templates were used to build the antibody model.
* All six loop templates have the same lengths as their corresponding target loops.
* CDR-L1 has a mismatch in canonical structure classification despite matching loop length.
* CDR-L2 has the lowest sequence identity to its selected loop template.
* CDR-L3 is the longest loop in the LYRA model.
* Several negatively charged residues are present in the CDRs, especially in the light chain.
* Template similarity does not guarantee structural accuracy, particularly in flexible loops.

== References ==

* [https://services.healthtech.dtu.dk/services/LYRA-1.0/ LYRA]
* [https://www.imgt.org/ IMGT]
* [https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/abodybuilder2/ ABodyBuilder2]

Exercise:AntibodyStructurePrediction ans

2026-10-21T02:29:10Z

Carol: /* Q3. Sequence differences between target and templates */

__TOC__

= Antibody Structure Modelling — Answers =

''SI135 CLL antibody modelling exercise.''

== Part 1 — Antibody modelling with LYRA ==

=== Q1. Framework template selection ===

'''Answer:'''

LYRA selected the following framework templates:

{| class="wikitable"
! Chain
! Selected template
! Sequence identity
! BLOSUM score
|-
| Heavy (VH)
| '''3H42'''
| '''80.5%'''
| 514
|-
| Lambda light (VL)
| '''1W72'''
| '''85.7%'''
| 518
|}

Interestingly, 2UZI has a slightly higher sequence identity to the heavy chain (81.7%), but LYRA selects 3H42, which has a higher BLOSUM score and compatible canonical structures.

This illustrates that template selection depends on more than sequence identity alone.

=== Q2. Light-chain type ===

'''Answer:'''

The SI135 antibody has a '''lambda (λ) light chain'''.

Lambda is less common than kappa (κ) in human antibodies. Approximately 60% of human antibodies have kappa light chains and 40% have lambda light chains.

=== Q3. Sequence differences between target and templates ===

'''Answer:'''

The target antibody is relatively similar to the selected framework templates, but several substitutions are observed in both framework and CDR regions.

Some examples from the LYRA alignments include:

{| class="wikitable"
! Region
! Template → Target
! Interpretation
|-
| Heavy-chain framework
| K → R
| Conservative substitution between positively charged residues
|-
| CDR-H1
| S → T
| Conservative polar substitution
|-
| CDR-H2
| S → N
| Changes side-chain size and hydrogen-bonding possibilities
|-
| CDR-H3
| R → N
| Removes a positive charge
|-
| CDR-H3
| Q → T
| Changes side-chain size and hydrogen-bonding properties
|-
| CDR-L2
| D → Y
| Removes a negative charge and introduces a bulky aromatic residue
|-
| CDR-L3
| L → S
| Changes from hydrophobic to polar
|-
| CDR-L3
| S → D
| Introduces a negative charge
|-
| CDR-L3
| A → H
| Introduces a larger, potentially ionizable side chain
|-
| CDR-L3
| D → P
| Removes a negative charge and introduces a residue that restricts backbone conformation
|}

The CDR-L2 substitution D → Y and the CDR-L3 substitution D → P are particularly interesting because they involve substantial changes in physicochemical properties.

In the light-chain framework alignment, a D → K substitution near the C-terminus also reverses the side-chain charge.

'''Note:''' The impact of a substitution depends not only on its BLOSUM score but also on its structural location and whether the residue is buried or exposed.

=== Q4. CDR loop lengths and expected accuracy ===

'''Answer:'''

All six selected CDR templates have the same loop length as the corresponding SI135 target loops.

{| class="wikitable"
! Loop
! Template
! Template sequence
! Target sequence
! Length
|-
| H1
| 3H42
| GFTFSSY
| GFTFSTY
| 7
|-
| H2
| 3H42
| SSSS
| SSSN
| 4
|-
| H3
| 2NR6
| VRQMD
| VNTMD
| 5
|-
| L1
| 2G75
| GNNIGSKS
| GNNIGSKS
| 8
|-
| L2
| 1W72
| DDS
| YDN
| 3
|-
| L3
| 3GHB
| WDSGLSADW
| WDSGSDHPW
| 9
|}

''These are the loop-core lengths reported by LYRA. Other CDR definitions may include additional residues.''

Matching loop lengths generally make template-based modelling easier because no insertions or deletions are required in the loop core.

However, there is one interesting exception in the canonical structure classification.

For CDR-L1:

* '''Template canonical structure: 7'''
* '''Predicted canonical structure: 8'''

Thus, although L1 has the same sequence length and an identical loop-core sequence in the selected template, the predicted canonical structure class is different.

This introduces additional uncertainty in the loop conformation.

'''Note:''' Matching loop length does not guarantee that the target and template have the same three-dimensional conformation.

=== Q5. Number of templates and most dissimilar region ===

'''Answer:'''

LYRA uses '''five distinct PDB templates''' to construct the SI135 antibody model:

{| class="wikitable"
! PDB template
! Regions modelled
|-
| '''3H42'''
| Heavy-chain framework, H1, H2 and heavy/light chain packing
|-
| '''2NR6'''
| H3
|-
| '''1W72'''
| Light-chain framework and L2
|-
| '''2G75'''
| L1
|-
| '''3GHB'''
| L3
|}

A total of eight framework/loop regions are modelled, but several use the same structural template.

The sequence identity within each selected loop core is:

{| class="wikitable"
! Loop
! Identical residues
! Identity
|-
| H1
| 6/7
| 85.7%
|-
| H2
| 3/4
| 75.0%
|-
| H3
| 3/5
| 60.0%
|-
| L1
| 8/8
| 100%
|-
| L2
| 1/3
| '''33.3%'''
|-
| L3
| 5/9
| 55.6%
|}

'''CDR-L2 has the lowest direct sequence identity''' between its selected loop template and the target.

CDR-L3 also contains several substitutions, including the potentially important D → P change.

The loop identities above are calculated directly from the sequences reported in the PDB annotations, not from the overall framework-template scores.

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Q6. Shape of the antigen-binding site ===

'''Expected answer:'''

The antigen-binding site is formed by the six CDR loops.

Its overall shape should be classified by displaying the predicted antibody as a molecular surface and examining the arrangement of these loops.

'''SI135 classification:''' To be confirmed from the PyMOL surface representation.

The LYRA sequence alignments do not establish whether the binding site is flat, pocket-like, groove-like or mountain-like.

The classification is qualitative, and students should justify their answer based on the model.

=== Q7. Longest and most protruding CDR ===

'''Answer:'''

According to LYRA, the six loop-core lengths are:

{| class="wikitable"
! Loop
! Length
|-
| H1
| 7
|-
| H2
| 4
|-
| H3
| 5
|-
| L1
| 8
|-
| L2
| 3
|-
| '''L3'''
| '''9'''
|}

The '''longest loop is CDR-L3'''.

This is notable because CDR-H3 is often the longest and most structurally variable CDR in human antibodies.

The loop that protrudes furthest should be determined from the three-dimensional structure, since length alone does not determine protrusion.

'''Most protruding loop:''' To be confirmed by visual inspection in PyMOL.

== Part 3 — Somatic mutations ==

=== Q8. Somatic mutations near the CDRs ===

'''Expected answer:'''

Somatic mutations must be identified by comparing the SI135 sequence with the closest germline immunoglobulin genes using IMGT/DomainGapAlign.

The target-template substitutions identified by LYRA are not necessarily somatic mutations, because the structural templates are unrelated antibody sequences rather than the germline ancestors of SI135.

Mutations located inside the CDRs or affecting the residues that support them are particularly interesting.

The most likely candidates to affect binding are substitutions that:

* Change the charge of an exposed residue.
* Introduce or remove aromatic residues.
* Affect hydrogen bonding.
* Change loop geometry or flexibility.

'''SI135-specific mutation list:''' Requires the IMGT germline alignment.

== Part 4 — Electrostatic properties ==

=== Q9. Charged CDR residues ===

'''Answer:'''

The LYRA loop definitions allow us to identify the following charged or potentially charged residues.

{| class="wikitable"
! Loop
! Sequence
! Relevant residues
|-
| H1
| GFTFSTY
| None of K, R, D or E
|-
| H2
| SSSN
| None of K, R, D or E
|-
| H3
| VNTMD
| D (negative)
|-
| L1
| GNNIGSKS
| K (positive)
|-
| L2
| YDN
| D (negative)
|-
| L3
| WDSGSDHPW
| Two D residues (negative); H may be protonated
|}

The principal charged residues within these CDR definitions are therefore:

* One aspartate in H3.
* One lysine in L1.
* One aspartate in L2.
* Two aspartates in L3.

The light-chain loops contain several negatively charged residues, which may contribute to negatively charged patches on the antigen-binding surface.

However, the actual electrostatic distribution depends on the spatial arrangement of these residues.

'''Residue numbering in the provided LYRA PDB:'''

{| class="wikitable"
! Loop
! Residue
! PDB position
|-
| H3
| Aspartate
| H101
|-
| L1
| Lysine
| L31
|-
| L2
| Aspartate
| L51
|-
| L3
| Aspartate
| L92
|-
| L3
| Aspartate
| L95A
|-
| L3
| Histidine
| L95B
|}

Students can highlight the main charged CDR residues using the following PyMOL commands:

<pre>
select charged_cdrs, (chain H and resi 101) or (chain L and resi 31+51+92+95A)
show sticks, charged_cdrs
color yellow, charged_cdrs
</pre>

''These residue numbers refer specifically to the supplied LYRA PDB file.''

=== Q10. Introducing a positive charge (optional) ===

'''Expected answer:'''

Replacing an exposed negatively charged residue with lysine or arginine should make its local electrostatic environment more positive.

For example, a D → K mutation at a solvent-exposed CDR position would reverse the formal side-chain charge.

After recalculating the electrostatic potential, the surface coloring near the mutation should change.

'''Actual changes:''' Require performing the mutation and recalculating the potential in PyMOL.

A change in electrostatic potential does not necessarily improve antigen binding.

== Part 5 — Comparing antibody modelling approaches ==

=== Q11. LYRA versus ABodyBuilder2 ===

'''Expected answer:'''

LYRA and ABodyBuilder2 use different approaches to structure prediction.

{| class="wikitable"
! LYRA
! ABodyBuilder2
|-
| Template-based homology modelling
| Deep-learning-based prediction
|-
| Uses experimentally determined structural templates
| Uses a trained neural network
|-
| Models framework and loops using selected templates
| Predicts the antibody structure from sequence
|-
| Depends on availability and suitability of structural templates
| Provides prediction-error estimates
|}

Both methods should produce a similar overall immunoglobulin fold.

Differences are expected to be more pronounced in variable loops, particularly where the sequence similarity to the LYRA template is low or the canonical structures differ.

For SI135, potentially interesting regions include:

* CDR-L1, because of the canonical structure mismatch.
* CDR-L2, because of its low sequence identity with the selected loop template.
* CDR-L3, because it contains several non-conservative substitutions.
* CDR-H3, because antibody H3 loops can have diverse conformations.

'''Actual differences between the two SI135 predictions:''' Require the ABodyBuilder2 model for structural superposition.

In PyMOL, the models can be superimposed using:

<pre>
align ABodyBuilder2, LYRA
</pre>

The object names should be adapted to the names used in the PyMOL session.

== Take-home messages ==

* The SI135 antibody has a lambda light chain.
* The selected framework templates show relatively high sequence identity: 80.5% for VH and 85.7% for VL.
* Five distinct structural templates were used to build the antibody model.
* All six loop templates have the same lengths as their corresponding target loops.
* CDR-L1 has a mismatch in canonical structure classification despite matching loop length.
* CDR-L2 has the lowest sequence identity to its selected loop template.
* CDR-L3 is the longest loop in the LYRA model.
* Several negatively charged residues are present in the CDRs, especially in the light chain.
* Template similarity does not guarantee structural accuracy, particularly in flexible loops.

== References ==

* [https://services.healthtech.dtu.dk/services/LYRA-1.0/ LYRA]
* [https://www.imgt.org/ IMGT]
* [https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/abodybuilder2/ ABodyBuilder2]

Exercise:AntibodyStructurePrediction ans

2026-10-21T02:28:58Z

Carol: /* Q4. CDR loop lengths and expected accuracy */

__TOC__

= Antibody Structure Modelling — Answers =

''SI135 CLL antibody modelling exercise.''

== Part 1 — Antibody modelling with LYRA ==

=== Q1. Framework template selection ===

'''Answer:'''

LYRA selected the following framework templates:

{| class="wikitable"
! Chain
! Selected template
! Sequence identity
! BLOSUM score
|-
| Heavy (VH)
| '''3H42'''
| '''80.5%'''
| 514
|-
| Lambda light (VL)
| '''1W72'''
| '''85.7%'''
| 518
|}

Interestingly, 2UZI has a slightly higher sequence identity to the heavy chain (81.7%), but LYRA selects 3H42, which has a higher BLOSUM score and compatible canonical structures.

This illustrates that template selection depends on more than sequence identity alone.

=== Q2. Light-chain type ===

'''Answer:'''

The SI135 antibody has a '''lambda (λ) light chain'''.

Lambda is less common than kappa (κ) in human antibodies. Approximately 60% of human antibodies have kappa light chains and 40% have lambda light chains.

=== Q3. Sequence differences between target and templates ===

'''Answer:'''

The target antibody is relatively similar to the selected framework templates, but several substitutions are observed in both framework and CDR regions.

Some examples from the LYRA alignments include:

{| class="wikitable"
! Region
! Template → Target
! Interpretation
|-
| Heavy-chain framework
| K → R
| Conservative substitution between positively charged residues
|-
| CDR-H1
| S → T
| Conservative polar substitution
|-
| CDR-H2
| S → N
| Changes side-chain size and hydrogen-bonding possibilities
|-
| CDR-H3
| R → N
| Removes a positive charge
|-
| CDR-H3
| Q → T
| Changes side-chain size and hydrogen-bonding properties
|-
| CDR-L2
| D → Y
| Removes a negative charge and introduces a bulky aromatic residue
|-
| CDR-L3
| L → S
| Changes from hydrophobic to polar
|-
| CDR-L3
| S → D
| Introduces a negative charge
|-
| CDR-L3
| A → H
| Introduces a larger, potentially ionizable side chain
|-
| CDR-L3
| D → P
| Removes a negative charge and introduces a residue that restricts backbone conformation
|}

The CDR-L2 substitution D → Y and the CDR-L3 substitution D → P are particularly interesting because they involve substantial changes in physicochemical properties.

In the light-chain framework alignment, a D → K substitution near the C-terminus also reverses the side-chain charge.

'''Observation:''' The impact of a substitution depends not only on its BLOSUM score but also on its structural location and whether the residue is buried or exposed.

=== Q4. CDR loop lengths and expected accuracy ===

'''Answer:'''

All six selected CDR templates have the same loop length as the corresponding SI135 target loops.

{| class="wikitable"
! Loop
! Template
! Template sequence
! Target sequence
! Length
|-
| H1
| 3H42
| GFTFSSY
| GFTFSTY
| 7
|-
| H2
| 3H42
| SSSS
| SSSN
| 4
|-
| H3
| 2NR6
| VRQMD
| VNTMD
| 5
|-
| L1
| 2G75
| GNNIGSKS
| GNNIGSKS
| 8
|-
| L2
| 1W72
| DDS
| YDN
| 3
|-
| L3
| 3GHB
| WDSGLSADW
| WDSGSDHPW
| 9
|}

''These are the loop-core lengths reported by LYRA. Other CDR definitions may include additional residues.''

Matching loop lengths generally make template-based modelling easier because no insertions or deletions are required in the loop core.

However, there is one interesting exception in the canonical structure classification.

For CDR-L1:

* '''Template canonical structure: 7'''
* '''Predicted canonical structure: 8'''

Thus, although L1 has the same sequence length and an identical loop-core sequence in the selected template, the predicted canonical structure class is different.

This introduces additional uncertainty in the loop conformation.

'''Note:''' Matching loop length does not guarantee that the target and template have the same three-dimensional conformation.

=== Q5. Number of templates and most dissimilar region ===

'''Answer:'''

LYRA uses '''five distinct PDB templates''' to construct the SI135 antibody model:

{| class="wikitable"
! PDB template
! Regions modelled
|-
| '''3H42'''
| Heavy-chain framework, H1, H2 and heavy/light chain packing
|-
| '''2NR6'''
| H3
|-
| '''1W72'''
| Light-chain framework and L2
|-
| '''2G75'''
| L1
|-
| '''3GHB'''
| L3
|}

A total of eight framework/loop regions are modelled, but several use the same structural template.

The sequence identity within each selected loop core is:

{| class="wikitable"
! Loop
! Identical residues
! Identity
|-
| H1
| 6/7
| 85.7%
|-
| H2
| 3/4
| 75.0%
|-
| H3
| 3/5
| 60.0%
|-
| L1
| 8/8
| 100%
|-
| L2
| 1/3
| '''33.3%'''
|-
| L3
| 5/9
| 55.6%
|}

'''CDR-L2 has the lowest direct sequence identity''' between its selected loop template and the target.

CDR-L3 also contains several substitutions, including the potentially important D → P change.

The loop identities above are calculated directly from the sequences reported in the PDB annotations, not from the overall framework-template scores.

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Q6. Shape of the antigen-binding site ===

'''Expected answer:'''

The antigen-binding site is formed by the six CDR loops.

Its overall shape should be classified by displaying the predicted antibody as a molecular surface and examining the arrangement of these loops.

'''SI135 classification:''' To be confirmed from the PyMOL surface representation.

The LYRA sequence alignments do not establish whether the binding site is flat, pocket-like, groove-like or mountain-like.

The classification is qualitative, and students should justify their answer based on the model.

=== Q7. Longest and most protruding CDR ===

'''Answer:'''

According to LYRA, the six loop-core lengths are:

{| class="wikitable"
! Loop
! Length
|-
| H1
| 7
|-
| H2
| 4
|-
| H3
| 5
|-
| L1
| 8
|-
| L2
| 3
|-
| '''L3'''
| '''9'''
|}

The '''longest loop is CDR-L3'''.

This is notable because CDR-H3 is often the longest and most structurally variable CDR in human antibodies.

The loop that protrudes furthest should be determined from the three-dimensional structure, since length alone does not determine protrusion.

'''Most protruding loop:''' To be confirmed by visual inspection in PyMOL.

== Part 3 — Somatic mutations ==

=== Q8. Somatic mutations near the CDRs ===

'''Expected answer:'''

Somatic mutations must be identified by comparing the SI135 sequence with the closest germline immunoglobulin genes using IMGT/DomainGapAlign.

The target-template substitutions identified by LYRA are not necessarily somatic mutations, because the structural templates are unrelated antibody sequences rather than the germline ancestors of SI135.

Mutations located inside the CDRs or affecting the residues that support them are particularly interesting.

The most likely candidates to affect binding are substitutions that:

* Change the charge of an exposed residue.
* Introduce or remove aromatic residues.
* Affect hydrogen bonding.
* Change loop geometry or flexibility.

'''SI135-specific mutation list:''' Requires the IMGT germline alignment.

== Part 4 — Electrostatic properties ==

=== Q9. Charged CDR residues ===

'''Answer:'''

The LYRA loop definitions allow us to identify the following charged or potentially charged residues.

{| class="wikitable"
! Loop
! Sequence
! Relevant residues
|-
| H1
| GFTFSTY
| None of K, R, D or E
|-
| H2
| SSSN
| None of K, R, D or E
|-
| H3
| VNTMD
| D (negative)
|-
| L1
| GNNIGSKS
| K (positive)
|-
| L2
| YDN
| D (negative)
|-
| L3
| WDSGSDHPW
| Two D residues (negative); H may be protonated
|}

The principal charged residues within these CDR definitions are therefore:

* One aspartate in H3.
* One lysine in L1.
* One aspartate in L2.
* Two aspartates in L3.

The light-chain loops contain several negatively charged residues, which may contribute to negatively charged patches on the antigen-binding surface.

However, the actual electrostatic distribution depends on the spatial arrangement of these residues.

'''Residue numbering in the provided LYRA PDB:'''

{| class="wikitable"
! Loop
! Residue
! PDB position
|-
| H3
| Aspartate
| H101
|-
| L1
| Lysine
| L31
|-
| L2
| Aspartate
| L51
|-
| L3
| Aspartate
| L92
|-
| L3
| Aspartate
| L95A
|-
| L3
| Histidine
| L95B
|}

Students can highlight the main charged CDR residues using the following PyMOL commands:

<pre>
select charged_cdrs, (chain H and resi 101) or (chain L and resi 31+51+92+95A)
show sticks, charged_cdrs
color yellow, charged_cdrs
</pre>

''These residue numbers refer specifically to the supplied LYRA PDB file.''

=== Q10. Introducing a positive charge (optional) ===

'''Expected answer:'''

Replacing an exposed negatively charged residue with lysine or arginine should make its local electrostatic environment more positive.

For example, a D → K mutation at a solvent-exposed CDR position would reverse the formal side-chain charge.

After recalculating the electrostatic potential, the surface coloring near the mutation should change.

'''Actual changes:''' Require performing the mutation and recalculating the potential in PyMOL.

A change in electrostatic potential does not necessarily improve antigen binding.

== Part 5 — Comparing antibody modelling approaches ==

=== Q11. LYRA versus ABodyBuilder2 ===

'''Expected answer:'''

LYRA and ABodyBuilder2 use different approaches to structure prediction.

{| class="wikitable"
! LYRA
! ABodyBuilder2
|-
| Template-based homology modelling
| Deep-learning-based prediction
|-
| Uses experimentally determined structural templates
| Uses a trained neural network
|-
| Models framework and loops using selected templates
| Predicts the antibody structure from sequence
|-
| Depends on availability and suitability of structural templates
| Provides prediction-error estimates
|}

Both methods should produce a similar overall immunoglobulin fold.

Differences are expected to be more pronounced in variable loops, particularly where the sequence similarity to the LYRA template is low or the canonical structures differ.

For SI135, potentially interesting regions include:

* CDR-L1, because of the canonical structure mismatch.
* CDR-L2, because of its low sequence identity with the selected loop template.
* CDR-L3, because it contains several non-conservative substitutions.
* CDR-H3, because antibody H3 loops can have diverse conformations.

'''Actual differences between the two SI135 predictions:''' Require the ABodyBuilder2 model for structural superposition.

In PyMOL, the models can be superimposed using:

<pre>
align ABodyBuilder2, LYRA
</pre>

The object names should be adapted to the names used in the PyMOL session.

== Take-home messages ==

* The SI135 antibody has a lambda light chain.
* The selected framework templates show relatively high sequence identity: 80.5% for VH and 85.7% for VL.
* Five distinct structural templates were used to build the antibody model.
* All six loop templates have the same lengths as their corresponding target loops.
* CDR-L1 has a mismatch in canonical structure classification despite matching loop length.
* CDR-L2 has the lowest sequence identity to its selected loop template.
* CDR-L3 is the longest loop in the LYRA model.
* Several negatively charged residues are present in the CDRs, especially in the light chain.
* Template similarity does not guarantee structural accuracy, particularly in flexible loops.

== References ==

* [https://services.healthtech.dtu.dk/services/LYRA-1.0/ LYRA]
* [https://www.imgt.org/ IMGT]
* [https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/abodybuilder2/ ABodyBuilder2]

Exercise:AntibodyStructurePrediction ans

2026-10-21T02:28:33Z

Carol: /* Q3. Sequence differences between target and templates */

__TOC__

= Antibody Structure Modelling — Answers =

''SI135 CLL antibody modelling exercise.''

== Part 1 — Antibody modelling with LYRA ==

=== Q1. Framework template selection ===

'''Answer:'''

LYRA selected the following framework templates:

{| class="wikitable"
! Chain
! Selected template
! Sequence identity
! BLOSUM score
|-
| Heavy (VH)
| '''3H42'''
| '''80.5%'''
| 514
|-
| Lambda light (VL)
| '''1W72'''
| '''85.7%'''
| 518
|}

Interestingly, 2UZI has a slightly higher sequence identity to the heavy chain (81.7%), but LYRA selects 3H42, which has a higher BLOSUM score and compatible canonical structures.

This illustrates that template selection depends on more than sequence identity alone.

=== Q2. Light-chain type ===

'''Answer:'''

The SI135 antibody has a '''lambda (λ) light chain'''.

Lambda is less common than kappa (κ) in human antibodies. Approximately 60% of human antibodies have kappa light chains and 40% have lambda light chains.

=== Q3. Sequence differences between target and templates ===

'''Answer:'''

The target antibody is relatively similar to the selected framework templates, but several substitutions are observed in both framework and CDR regions.

Some examples from the LYRA alignments include:

{| class="wikitable"
! Region
! Template → Target
! Interpretation
|-
| Heavy-chain framework
| K → R
| Conservative substitution between positively charged residues
|-
| CDR-H1
| S → T
| Conservative polar substitution
|-
| CDR-H2
| S → N
| Changes side-chain size and hydrogen-bonding possibilities
|-
| CDR-H3
| R → N
| Removes a positive charge
|-
| CDR-H3
| Q → T
| Changes side-chain size and hydrogen-bonding properties
|-
| CDR-L2
| D → Y
| Removes a negative charge and introduces a bulky aromatic residue
|-
| CDR-L3
| L → S
| Changes from hydrophobic to polar
|-
| CDR-L3
| S → D
| Introduces a negative charge
|-
| CDR-L3
| A → H
| Introduces a larger, potentially ionizable side chain
|-
| CDR-L3
| D → P
| Removes a negative charge and introduces a residue that restricts backbone conformation
|}

The CDR-L2 substitution D → Y and the CDR-L3 substitution D → P are particularly interesting because they involve substantial changes in physicochemical properties.

In the light-chain framework alignment, a D → K substitution near the C-terminus also reverses the side-chain charge.

'''Observation:''' The impact of a substitution depends not only on its BLOSUM score but also on its structural location and whether the residue is buried or exposed.

=== Q4. CDR loop lengths and expected accuracy ===

'''Answer:'''

All six selected CDR templates have the same loop length as the corresponding SI135 target loops.

{| class="wikitable"
! Loop
! Template
! Template sequence
! Target sequence
! Length
|-
| H1
| 3H42
| GFTFSSY
| GFTFSTY
| 7
|-
| H2
| 3H42
| SSSS
| SSSN
| 4
|-
| H3
| 2NR6
| VRQMD
| VNTMD
| 5
|-
| L1
| 2G75
| GNNIGSKS
| GNNIGSKS
| 8
|-
| L2
| 1W72
| DDS
| YDN
| 3
|-
| L3
| 3GHB
| WDSGLSADW
| WDSGSDHPW
| 9
|}

''These are the loop-core lengths reported by LYRA. Other CDR definitions may include additional residues.''

Matching loop lengths generally make template-based modelling easier because no insertions or deletions are required in the loop core.

However, there is one interesting exception in the canonical structure classification.

For CDR-L1:

* '''Template canonical structure: 7'''
* '''Predicted canonical structure: 8'''

Thus, although L1 has the same sequence length and an identical loop-core sequence in the selected template, the predicted canonical structure class is different.

This introduces additional uncertainty in the loop conformation.

'''Teaching point:''' Matching loop length does not guarantee that the target and template have the same three-dimensional conformation.

=== Q5. Number of templates and most dissimilar region ===

'''Answer:'''

LYRA uses '''five distinct PDB templates''' to construct the SI135 antibody model:

{| class="wikitable"
! PDB template
! Regions modelled
|-
| '''3H42'''
| Heavy-chain framework, H1, H2 and heavy/light chain packing
|-
| '''2NR6'''
| H3
|-
| '''1W72'''
| Light-chain framework and L2
|-
| '''2G75'''
| L1
|-
| '''3GHB'''
| L3
|}

A total of eight framework/loop regions are modelled, but several use the same structural template.

The sequence identity within each selected loop core is:

{| class="wikitable"
! Loop
! Identical residues
! Identity
|-
| H1
| 6/7
| 85.7%
|-
| H2
| 3/4
| 75.0%
|-
| H3
| 3/5
| 60.0%
|-
| L1
| 8/8
| 100%
|-
| L2
| 1/3
| '''33.3%'''
|-
| L3
| 5/9
| 55.6%
|}

'''CDR-L2 has the lowest direct sequence identity''' between its selected loop template and the target.

CDR-L3 also contains several substitutions, including the potentially important D → P change.

The loop identities above are calculated directly from the sequences reported in the PDB annotations, not from the overall framework-template scores.

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Q6. Shape of the antigen-binding site ===

'''Expected answer:'''

The antigen-binding site is formed by the six CDR loops.

Its overall shape should be classified by displaying the predicted antibody as a molecular surface and examining the arrangement of these loops.

'''SI135 classification:''' To be confirmed from the PyMOL surface representation.

The LYRA sequence alignments do not establish whether the binding site is flat, pocket-like, groove-like or mountain-like.

The classification is qualitative, and students should justify their answer based on the model.

=== Q7. Longest and most protruding CDR ===

'''Answer:'''

According to LYRA, the six loop-core lengths are:

{| class="wikitable"
! Loop
! Length
|-
| H1
| 7
|-
| H2
| 4
|-
| H3
| 5
|-
| L1
| 8
|-
| L2
| 3
|-
| '''L3'''
| '''9'''
|}

The '''longest loop is CDR-L3'''.

This is notable because CDR-H3 is often the longest and most structurally variable CDR in human antibodies.

The loop that protrudes furthest should be determined from the three-dimensional structure, since length alone does not determine protrusion.

'''Most protruding loop:''' To be confirmed by visual inspection in PyMOL.

== Part 3 — Somatic mutations ==

=== Q8. Somatic mutations near the CDRs ===

'''Expected answer:'''

Somatic mutations must be identified by comparing the SI135 sequence with the closest germline immunoglobulin genes using IMGT/DomainGapAlign.

The target-template substitutions identified by LYRA are not necessarily somatic mutations, because the structural templates are unrelated antibody sequences rather than the germline ancestors of SI135.

Mutations located inside the CDRs or affecting the residues that support them are particularly interesting.

The most likely candidates to affect binding are substitutions that:

* Change the charge of an exposed residue.
* Introduce or remove aromatic residues.
* Affect hydrogen bonding.
* Change loop geometry or flexibility.

'''SI135-specific mutation list:''' Requires the IMGT germline alignment.

== Part 4 — Electrostatic properties ==

=== Q9. Charged CDR residues ===

'''Answer:'''

The LYRA loop definitions allow us to identify the following charged or potentially charged residues.

{| class="wikitable"
! Loop
! Sequence
! Relevant residues
|-
| H1
| GFTFSTY
| None of K, R, D or E
|-
| H2
| SSSN
| None of K, R, D or E
|-
| H3
| VNTMD
| D (negative)
|-
| L1
| GNNIGSKS
| K (positive)
|-
| L2
| YDN
| D (negative)
|-
| L3
| WDSGSDHPW
| Two D residues (negative); H may be protonated
|}

The principal charged residues within these CDR definitions are therefore:

* One aspartate in H3.
* One lysine in L1.
* One aspartate in L2.
* Two aspartates in L3.

The light-chain loops contain several negatively charged residues, which may contribute to negatively charged patches on the antigen-binding surface.

However, the actual electrostatic distribution depends on the spatial arrangement of these residues.

'''Residue numbering in the provided LYRA PDB:'''

{| class="wikitable"
! Loop
! Residue
! PDB position
|-
| H3
| Aspartate
| H101
|-
| L1
| Lysine
| L31
|-
| L2
| Aspartate
| L51
|-
| L3
| Aspartate
| L92
|-
| L3
| Aspartate
| L95A
|-
| L3
| Histidine
| L95B
|}

Students can highlight the main charged CDR residues using the following PyMOL commands:

<pre>
select charged_cdrs, (chain H and resi 101) or (chain L and resi 31+51+92+95A)
show sticks, charged_cdrs
color yellow, charged_cdrs
</pre>

''These residue numbers refer specifically to the supplied LYRA PDB file.''

=== Q10. Introducing a positive charge (optional) ===

'''Expected answer:'''

Replacing an exposed negatively charged residue with lysine or arginine should make its local electrostatic environment more positive.

For example, a D → K mutation at a solvent-exposed CDR position would reverse the formal side-chain charge.

After recalculating the electrostatic potential, the surface coloring near the mutation should change.

'''Actual changes:''' Require performing the mutation and recalculating the potential in PyMOL.

A change in electrostatic potential does not necessarily improve antigen binding.

== Part 5 — Comparing antibody modelling approaches ==

=== Q11. LYRA versus ABodyBuilder2 ===

'''Expected answer:'''

LYRA and ABodyBuilder2 use different approaches to structure prediction.

{| class="wikitable"
! LYRA
! ABodyBuilder2
|-
| Template-based homology modelling
| Deep-learning-based prediction
|-
| Uses experimentally determined structural templates
| Uses a trained neural network
|-
| Models framework and loops using selected templates
| Predicts the antibody structure from sequence
|-
| Depends on availability and suitability of structural templates
| Provides prediction-error estimates
|}

Both methods should produce a similar overall immunoglobulin fold.

Differences are expected to be more pronounced in variable loops, particularly where the sequence similarity to the LYRA template is low or the canonical structures differ.

For SI135, potentially interesting regions include:

* CDR-L1, because of the canonical structure mismatch.
* CDR-L2, because of its low sequence identity with the selected loop template.
* CDR-L3, because it contains several non-conservative substitutions.
* CDR-H3, because antibody H3 loops can have diverse conformations.

'''Actual differences between the two SI135 predictions:''' Require the ABodyBuilder2 model for structural superposition.

In PyMOL, the models can be superimposed using:

<pre>
align ABodyBuilder2, LYRA
</pre>

The object names should be adapted to the names used in the PyMOL session.

== Take-home messages ==

* The SI135 antibody has a lambda light chain.
* The selected framework templates show relatively high sequence identity: 80.5% for VH and 85.7% for VL.
* Five distinct structural templates were used to build the antibody model.
* All six loop templates have the same lengths as their corresponding target loops.
* CDR-L1 has a mismatch in canonical structure classification despite matching loop length.
* CDR-L2 has the lowest sequence identity to its selected loop template.
* CDR-L3 is the longest loop in the LYRA model.
* Several negatively charged residues are present in the CDRs, especially in the light chain.
* Template similarity does not guarantee structural accuracy, particularly in flexible loops.

== References ==

* [https://services.healthtech.dtu.dk/services/LYRA-1.0/ LYRA]
* [https://www.imgt.org/ IMGT]
* [https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/abodybuilder2/ ABodyBuilder2]

Exercise:AntibodyStructurePrediction ans

2026-10-21T02:27:22Z

Carol:

__TOC__

= Antibody Structure Modelling — Answers =

''SI135 CLL antibody modelling exercise.''

== Part 1 — Antibody modelling with LYRA ==

=== Q1. Framework template selection ===

'''Answer:'''

LYRA selected the following framework templates:

{| class="wikitable"
! Chain
! Selected template
! Sequence identity
! BLOSUM score
|-
| Heavy (VH)
| '''3H42'''
| '''80.5%'''
| 514
|-
| Lambda light (VL)
| '''1W72'''
| '''85.7%'''
| 518
|}

Interestingly, 2UZI has a slightly higher sequence identity to the heavy chain (81.7%), but LYRA selects 3H42, which has a higher BLOSUM score and compatible canonical structures.

This illustrates that template selection depends on more than sequence identity alone.

=== Q2. Light-chain type ===

'''Answer:'''

The SI135 antibody has a '''lambda (λ) light chain'''.

Lambda is less common than kappa (κ) in human antibodies. Approximately 60% of human antibodies have kappa light chains and 40% have lambda light chains.

=== Q3. Sequence differences between target and templates ===

'''Answer:'''

The target antibody is relatively similar to the selected framework templates, but several substitutions are observed in both framework and CDR regions.

Some examples from the LYRA alignments include:

{| class="wikitable"
! Region
! Template → Target
! Interpretation
|-
| Heavy-chain framework
| K → R
| Conservative substitution between positively charged residues
|-
| CDR-H1
| S → T
| Conservative polar substitution
|-
| CDR-H2
| S → N
| Changes side-chain size and hydrogen-bonding possibilities
|-
| CDR-H3
| R → N
| Removes a positive charge
|-
| CDR-H3
| Q → T
| Changes side-chain size and hydrogen-bonding properties
|-
| CDR-L2
| D → Y
| Removes a negative charge and introduces a bulky aromatic residue
|-
| CDR-L3
| L → S
| Changes from hydrophobic to polar
|-
| CDR-L3
| S → D
| Introduces a negative charge
|-
| CDR-L3
| A → H
| Introduces a larger, potentially ionizable side chain
|-
| CDR-L3
| D → P
| Removes a negative charge and introduces a residue that restricts backbone conformation
|}

The CDR-L2 substitution D → Y and the CDR-L3 substitution D → P are particularly interesting because they involve substantial changes in physicochemical properties.

In the light-chain framework alignment, a D → K substitution near the C-terminus also reverses the side-chain charge.

'''Teaching point:''' The impact of a substitution depends not only on its BLOSUM score but also on its structural location and whether the residue is buried or exposed.

=== Q4. CDR loop lengths and expected accuracy ===

'''Answer:'''

All six selected CDR templates have the same loop length as the corresponding SI135 target loops.

{| class="wikitable"
! Loop
! Template
! Template sequence
! Target sequence
! Length
|-
| H1
| 3H42
| GFTFSSY
| GFTFSTY
| 7
|-
| H2
| 3H42
| SSSS
| SSSN
| 4
|-
| H3
| 2NR6
| VRQMD
| VNTMD
| 5
|-
| L1
| 2G75
| GNNIGSKS
| GNNIGSKS
| 8
|-
| L2
| 1W72
| DDS
| YDN
| 3
|-
| L3
| 3GHB
| WDSGLSADW
| WDSGSDHPW
| 9
|}

''These are the loop-core lengths reported by LYRA. Other CDR definitions may include additional residues.''

Matching loop lengths generally make template-based modelling easier because no insertions or deletions are required in the loop core.

However, there is one interesting exception in the canonical structure classification.

For CDR-L1:

* '''Template canonical structure: 7'''
* '''Predicted canonical structure: 8'''

Thus, although L1 has the same sequence length and an identical loop-core sequence in the selected template, the predicted canonical structure class is different.

This introduces additional uncertainty in the loop conformation.

'''Teaching point:''' Matching loop length does not guarantee that the target and template have the same three-dimensional conformation.

=== Q5. Number of templates and most dissimilar region ===

'''Answer:'''

LYRA uses '''five distinct PDB templates''' to construct the SI135 antibody model:

{| class="wikitable"
! PDB template
! Regions modelled
|-
| '''3H42'''
| Heavy-chain framework, H1, H2 and heavy/light chain packing
|-
| '''2NR6'''
| H3
|-
| '''1W72'''
| Light-chain framework and L2
|-
| '''2G75'''
| L1
|-
| '''3GHB'''
| L3
|}

A total of eight framework/loop regions are modelled, but several use the same structural template.

The sequence identity within each selected loop core is:

{| class="wikitable"
! Loop
! Identical residues
! Identity
|-
| H1
| 6/7
| 85.7%
|-
| H2
| 3/4
| 75.0%
|-
| H3
| 3/5
| 60.0%
|-
| L1
| 8/8
| 100%
|-
| L2
| 1/3
| '''33.3%'''
|-
| L3
| 5/9
| 55.6%
|}

'''CDR-L2 has the lowest direct sequence identity''' between its selected loop template and the target.

CDR-L3 also contains several substitutions, including the potentially important D → P change.

The loop identities above are calculated directly from the sequences reported in the PDB annotations, not from the overall framework-template scores.

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Q6. Shape of the antigen-binding site ===

'''Expected answer:'''

The antigen-binding site is formed by the six CDR loops.

Its overall shape should be classified by displaying the predicted antibody as a molecular surface and examining the arrangement of these loops.

'''SI135 classification:''' To be confirmed from the PyMOL surface representation.

The LYRA sequence alignments do not establish whether the binding site is flat, pocket-like, groove-like or mountain-like.

The classification is qualitative, and students should justify their answer based on the model.

=== Q7. Longest and most protruding CDR ===

'''Answer:'''

According to LYRA, the six loop-core lengths are:

{| class="wikitable"
! Loop
! Length
|-
| H1
| 7
|-
| H2
| 4
|-
| H3
| 5
|-
| L1
| 8
|-
| L2
| 3
|-
| '''L3'''
| '''9'''
|}

The '''longest loop is CDR-L3'''.

This is notable because CDR-H3 is often the longest and most structurally variable CDR in human antibodies.

The loop that protrudes furthest should be determined from the three-dimensional structure, since length alone does not determine protrusion.

'''Most protruding loop:''' To be confirmed by visual inspection in PyMOL.

== Part 3 — Somatic mutations ==

=== Q8. Somatic mutations near the CDRs ===

'''Expected answer:'''

Somatic mutations must be identified by comparing the SI135 sequence with the closest germline immunoglobulin genes using IMGT/DomainGapAlign.

The target-template substitutions identified by LYRA are not necessarily somatic mutations, because the structural templates are unrelated antibody sequences rather than the germline ancestors of SI135.

Mutations located inside the CDRs or affecting the residues that support them are particularly interesting.

The most likely candidates to affect binding are substitutions that:

* Change the charge of an exposed residue.
* Introduce or remove aromatic residues.
* Affect hydrogen bonding.
* Change loop geometry or flexibility.

'''SI135-specific mutation list:''' Requires the IMGT germline alignment.

== Part 4 — Electrostatic properties ==

=== Q9. Charged CDR residues ===

'''Answer:'''

The LYRA loop definitions allow us to identify the following charged or potentially charged residues.

{| class="wikitable"
! Loop
! Sequence
! Relevant residues
|-
| H1
| GFTFSTY
| None of K, R, D or E
|-
| H2
| SSSN
| None of K, R, D or E
|-
| H3
| VNTMD
| D (negative)
|-
| L1
| GNNIGSKS
| K (positive)
|-
| L2
| YDN
| D (negative)
|-
| L3
| WDSGSDHPW
| Two D residues (negative); H may be protonated
|}

The principal charged residues within these CDR definitions are therefore:

* One aspartate in H3.
* One lysine in L1.
* One aspartate in L2.
* Two aspartates in L3.

The light-chain loops contain several negatively charged residues, which may contribute to negatively charged patches on the antigen-binding surface.

However, the actual electrostatic distribution depends on the spatial arrangement of these residues.

'''Residue numbering in the provided LYRA PDB:'''

{| class="wikitable"
! Loop
! Residue
! PDB position
|-
| H3
| Aspartate
| H101
|-
| L1
| Lysine
| L31
|-
| L2
| Aspartate
| L51
|-
| L3
| Aspartate
| L92
|-
| L3
| Aspartate
| L95A
|-
| L3
| Histidine
| L95B
|}

Students can highlight the main charged CDR residues using the following PyMOL commands:

<pre>
select charged_cdrs, (chain H and resi 101) or (chain L and resi 31+51+92+95A)
show sticks, charged_cdrs
color yellow, charged_cdrs
</pre>

''These residue numbers refer specifically to the supplied LYRA PDB file.''

=== Q10. Introducing a positive charge (optional) ===

'''Expected answer:'''

Replacing an exposed negatively charged residue with lysine or arginine should make its local electrostatic environment more positive.

For example, a D → K mutation at a solvent-exposed CDR position would reverse the formal side-chain charge.

After recalculating the electrostatic potential, the surface coloring near the mutation should change.

'''Actual changes:''' Require performing the mutation and recalculating the potential in PyMOL.

A change in electrostatic potential does not necessarily improve antigen binding.

== Part 5 — Comparing antibody modelling approaches ==

=== Q11. LYRA versus ABodyBuilder2 ===

'''Expected answer:'''

LYRA and ABodyBuilder2 use different approaches to structure prediction.

{| class="wikitable"
! LYRA
! ABodyBuilder2
|-
| Template-based homology modelling
| Deep-learning-based prediction
|-
| Uses experimentally determined structural templates
| Uses a trained neural network
|-
| Models framework and loops using selected templates
| Predicts the antibody structure from sequence
|-
| Depends on availability and suitability of structural templates
| Provides prediction-error estimates
|}

Both methods should produce a similar overall immunoglobulin fold.

Differences are expected to be more pronounced in variable loops, particularly where the sequence similarity to the LYRA template is low or the canonical structures differ.

For SI135, potentially interesting regions include:

* CDR-L1, because of the canonical structure mismatch.
* CDR-L2, because of its low sequence identity with the selected loop template.
* CDR-L3, because it contains several non-conservative substitutions.
* CDR-H3, because antibody H3 loops can have diverse conformations.

'''Actual differences between the two SI135 predictions:''' Require the ABodyBuilder2 model for structural superposition.

In PyMOL, the models can be superimposed using:

<pre>
align ABodyBuilder2, LYRA
</pre>

The object names should be adapted to the names used in the PyMOL session.

== Take-home messages ==

* The SI135 antibody has a lambda light chain.
* The selected framework templates show relatively high sequence identity: 80.5% for VH and 85.7% for VL.
* Five distinct structural templates were used to build the antibody model.
* All six loop templates have the same lengths as their corresponding target loops.
* CDR-L1 has a mismatch in canonical structure classification despite matching loop length.
* CDR-L2 has the lowest sequence identity to its selected loop template.
* CDR-L3 is the longest loop in the LYRA model.
* Several negatively charged residues are present in the CDRs, especially in the light chain.
* Template similarity does not guarantee structural accuracy, particularly in flexible loops.

== References ==

* [https://services.healthtech.dtu.dk/services/LYRA-1.0/ LYRA]
* [https://www.imgt.org/ IMGT]
* [https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/abodybuilder2/ ABodyBuilder2]

Exercise:AntibodyStructurePrediction

2026-10-21T02:25:34Z

Carol: /* Visualizing the antigen-binding site */

= Antibody Structure Modelling: From Sequence to 3D Structure =

__TOC__

== Introduction: ==

Antibodies recognize antigens through their variable domains, which contain six complementarity-determining regions (CDRs): three in the heavy chain and three in the light chain. These loops form the antigen-binding site and are supported by more conserved framework (FW) regions.

Predicting the three-dimensional structure of antibodies helps us understand how their sequence, shape, and chemical properties may influence antigen recognition. Different computational approaches, from traditional template-based modelling to modern deep-learning methods, can be used to predict antibody structures.

In this exercise, you will model an antibody from a patient with chronic lymphocytic leukemia (CLL), explore its structure using PyMOL, and investigate how sequence variation and structural features may influence antigen binding.

== Learning objectives ==

By the end of this exercise, you should be able to:

* Identify the framework and CDR regions of an antibody.
* Understand the basic principles of template-based antibody modelling.
* Generate and visualize a three-dimensional antibody model using computational tools.
* Explore how CDR sequence, structure, and electrostatic properties may influence antigen recognition.
* Identify somatic mutations and discuss their potential effects on antibody binding.
* Compare antibody structures predicted using traditional template-based and modern deep-learning approaches.

== Antibody sequences ==

We will use the variable regions of the SI135 antibody.

'''Light chain (VL):'''

<pre>
>SI135_VL
SYVVTQPPSVSVAPGKTARITCGGNNIGSKSVHWYQQKPGQAPVLVIYYDNDRPSGIPERFSGSNSGNTATLTISRVEAGDEADYYCQVWDSGSDHPWVFGGGTKLTVL
</pre>

'''Heavy chain (VH):'''

<pre>
>SI135_VH
EVQLVESGGGLVRPGGSLRLSCAASGFTFSTYSMNWVRQAPGKGLEWVSSISSSSNSIYCADSVKGRFTISRDNANNSLYMQMNSLRAEDTAVYYCARDVNTMDVWGQGILVTVLL
</pre>

== Part 1 — Identify the antibody regions ==

Go to the [https://services.healthtech.dtu.dk/services/LYRA-1.0/ Lyra] web page.

''Lyra is a tool for predicting antibody and TCR structures. It is possible to use Lyra in automated or expert mode.''

Insert the light and heavy chains in the proper text area and submit. After a few seconds you will arrive to the output page, where the alignment and all the templates are displayed.

Note that for each Region of each chain of the antibody a Template has been found.

{| class="wikitable"
! Region
! Templates
|-
| Framework
| ...
|-
| Loop 1 (CDR1)
| ...
|-
| Loop 2 (CDR2)
| ...
|-
| Loop 3 (CDR3)
| ...
|}

Note that after the Regions there is an extra template used for packing the whole structure.

Explore the output page and the different tabs - check the PDB file, Log, and manual selection ones to find this info.

Now answer the following questions:

'''Q1: what is the sequence similarity of the best H and L templates for the framework (FW)?'''

'''Q2: What is the light chain type? Is it the most common in humans?'''

=== Sequence alignment and CDR modelling ===

Focus on the sequence alignment used to generate the model. You can find it in one of the tabs in Lyra website, as well as in the PDB file itself.

'''Q3: where are the differences between target and template located? Which are the most drastic that could affect the modeling process (e.g. G->W)?'''

''hint: if you don't remeber the amino acids, you can use the Blosum62 matrix to estimate the expected effect of a substitution: the lower the number, the larger the expected impact. For sake of time, you only need to give a qualitative answer.''

'''Q4: Do the loops/CDR have the same length in target and template? What does it mean, in terms of expected accuracy?'''

=== Selection of structural templates ===

Now, look at the templates automatically selected by Lyra to build the final model.

'''Q5: How many templates were used? Which is the region for which the template is more dissimilar to the target sequence?'''

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Visualizing the antigen-binding site (ABS) ===

Download the pdb file and open it with pymol. There is a small "S" bottom right, it shows/hides the sequence. Show the whole protein as cartoon and color by chain. Select and color in red all the CDRs. (the selected residues appear as the "sele" item in the menu). The CDR definition can be seen in the alignment generated by Lyra.

'''Q6: What is the general shape of the ABS: flat, a pocket, a mountain or a groove?''' ''It can help to show the antibody as surface''

=== CDR loop structure ===

'''Q7: Which loop is the longest and protruding the most? Is this the most common case?'''

== Part 3 — Somatic mutations ==

=== Identifying mutations in the antibody sequence ===

Use the [http://www.imgt.org/3Dstructure-DB/cgi/DomainGapAlign.cgi Domain Gap Align tool] identify the somatic mutations in the V and J regions. You need to use the amino acid sequence and select human as the source organism. In PyMol, show the mutations as sticks (main chain and side chain) and color them green.

'''Q8: Which Mutations are located inside/near the CDRs? Which are the ones that are most likely to have an impact on the binding of the antibody?''' ''A qualitative answer is sufficient''

== Part 4 — Electrostatic properties of the antigen-binding site ==

=== Visualizing the electrostatic potential ===

Generate the electrostatic potentials of the Ig ('''A'''ction -> generate -> vacuum electrostatic -> protein contact potentials).

'''Q9: Identify the residues in the CDRs that mostly contribute to the charge distribution.'''

''highlight them in the sequence and look for charged residues such as K, R, D, E''

=== Modifying the electrostatic properties (optional) ===

'''Q10: (optional) Mutate some residues in order to have a positive charge in the middle of the antigen binding site, then recalculate the electrostatic potentials. Can you see the changes?'''

== Part 5 — Comparing antibody modelling approaches ==

=== LYRA versus ABodyBuilder2 ===

'''Q11: Go to the AbodyBuilder 2 site and make a new structure prediction. Download the PDB file and compare with the model generated by LYRA. Can you see some differences?'''

'''Finished!'''

Exercise:AntibodyStructurePrediction

2026-10-21T02:24:12Z

Carol: /* Part 1 — Identify the antibody regions */

= Antibody Structure Modelling: From Sequence to 3D Structure =

__TOC__

== Introduction: ==

Antibodies recognize antigens through their variable domains, which contain six complementarity-determining regions (CDRs): three in the heavy chain and three in the light chain. These loops form the antigen-binding site and are supported by more conserved framework (FW) regions.

Predicting the three-dimensional structure of antibodies helps us understand how their sequence, shape, and chemical properties may influence antigen recognition. Different computational approaches, from traditional template-based modelling to modern deep-learning methods, can be used to predict antibody structures.

In this exercise, you will model an antibody from a patient with chronic lymphocytic leukemia (CLL), explore its structure using PyMOL, and investigate how sequence variation and structural features may influence antigen binding.

== Learning objectives ==

By the end of this exercise, you should be able to:

* Identify the framework and CDR regions of an antibody.
* Understand the basic principles of template-based antibody modelling.
* Generate and visualize a three-dimensional antibody model using computational tools.
* Explore how CDR sequence, structure, and electrostatic properties may influence antigen recognition.
* Identify somatic mutations and discuss their potential effects on antibody binding.
* Compare antibody structures predicted using traditional template-based and modern deep-learning approaches.

== Antibody sequences ==

We will use the variable regions of the SI135 antibody.

'''Light chain (VL):'''

<pre>
>SI135_VL
SYVVTQPPSVSVAPGKTARITCGGNNIGSKSVHWYQQKPGQAPVLVIYYDNDRPSGIPERFSGSNSGNTATLTISRVEAGDEADYYCQVWDSGSDHPWVFGGGTKLTVL
</pre>

'''Heavy chain (VH):'''

<pre>
>SI135_VH
EVQLVESGGGLVRPGGSLRLSCAASGFTFSTYSMNWVRQAPGKGLEWVSSISSSSNSIYCADSVKGRFTISRDNANNSLYMQMNSLRAEDTAVYYCARDVNTMDVWGQGILVTVLL
</pre>

== Part 1 — Identify the antibody regions ==

Go to the [https://services.healthtech.dtu.dk/services/LYRA-1.0/ Lyra] web page.

''Lyra is a tool for predicting antibody and TCR structures. It is possible to use Lyra in automated or expert mode.''

Insert the light and heavy chains in the proper text area and submit. After a few seconds you will arrive to the output page, where the alignment and all the templates are displayed.

Note that for each Region of each chain of the antibody a Template has been found.

{| class="wikitable"
! Region
! Templates
|-
| Framework
| ...
|-
| Loop 1 (CDR1)
| ...
|-
| Loop 2 (CDR2)
| ...
|-
| Loop 3 (CDR3)
| ...
|}

Note that after the Regions there is an extra template used for packing the whole structure.

Explore the output page and the different tabs - check the PDB file, Log, and manual selection ones to find this info.

Now answer the following questions:

'''Q1: what is the sequence similarity of the best H and L templates for the framework (FW)?'''

'''Q2: What is the light chain type? Is it the most common in humans?'''

=== Sequence alignment and CDR modelling ===

Focus on the sequence alignment used to generate the model. You can find it in one of the tabs in Lyra website, as well as in the PDB file itself.

'''Q3: where are the differences between target and template located? Which are the most drastic that could affect the modeling process (e.g. G->W)?'''

''hint: if you don't remeber the amino acids, you can use the Blosum62 matrix to estimate the expected effect of a substitution: the lower the number, the larger the expected impact. For sake of time, you only need to give a qualitative answer.''

'''Q4: Do the loops/CDR have the same length in target and template? What does it mean, in terms of expected accuracy?'''

=== Selection of structural templates ===

Now, look at the templates automatically selected by Lyra to build the final model.

'''Q5: How many templates were used? Which is the region for which the template is more dissimilar to the target sequence?'''

== Part 2 — Exploring the antibody structure in PyMOL ==

=== Visualizing the antigen-binding site ===

Download the pdb file and open it with pymol. There is a small "S" bottom right, it shows/hides the sequence. Show the whole protein as cartoon and color by chain. Select and color in red all the CDRs. (the selected residues appear as the "sele" item in the menu). The CDR definition can be seen in the alignment generated by Lyra.

'''Q6: What is the general shape of the ABS: flat, a pocket, a mountain or a groove?''' ''It can help to show the antibody as surface''

=== CDR loop structure ===

'''Q7: Which loop is the longest and protruding the most? Is this the most common case?'''

== Part 3 — Somatic mutations ==

=== Identifying mutations in the antibody sequence ===

Use the [http://www.imgt.org/3Dstructure-DB/cgi/DomainGapAlign.cgi Domain Gap Align tool] identify the somatic mutations in the V and J regions. You need to use the amino acid sequence and select human as the source organism. In PyMol, show the mutations as sticks (main chain and side chain) and color them green.

'''Q8: Which Mutations are located inside/near the CDRs? Which are the ones that are most likely to have an impact on the binding of the antibody?''' ''A qualitative answer is sufficient''

== Part 4 — Electrostatic properties of the antigen-binding site ==

=== Visualizing the electrostatic potential ===

Generate the electrostatic potentials of the Ig ('''A'''ction -> generate -> vacuum electrostatic -> protein contact potentials).

'''Q9: Identify the residues in the CDRs that mostly contribute to the charge distribution.'''

''highlight them in the sequence and look for charged residues such as K, R, D, E''

=== Modifying the electrostatic properties (optional) ===

'''Q10: (optional) Mutate some residues in order to have a positive charge in the middle of the antigen binding site, then recalculate the electrostatic potentials. Can you see the changes?'''

== Part 5 — Comparing antibody modelling approaches ==

=== LYRA versus ABodyBuilder2 ===

'''Q11: Go to the AbodyBuilder 2 site and make a new structure prediction. Download the PDB file and compare with the model generated by LYRA. Can you see some differences?'''

'''Finished!'''

Exercise:AntibodyStructurePrediction ans

2026-10-21T02:20:36Z

Carol:

__TOC__

= Antibody Structure Modelling Answers =

'''Case study:''' SI135 antibody from chronic lymphocytic leukemia (CLL)

'''LYRA prediction:''' [https://services.healthtech.dtu.dk/cgi-bin/webface2.cgi?jobid=6AD8166000282F67166ED625&wait=20 View the original modelling job]

== Part 1 — Antibody modelling with LYRA ==

=== Q1. Framework template sequence similarity ===

'''Q1: what is the sequence similarity of the best H and L templates for the framework (FW)?'''

'''Answer:'''

{| class="wikitable"
! Region
! Selected template
! Sequence identity
|-
| Heavy-chain framework (VH)
| [TO VERIFY]
| [TO VERIFY]
|-
| Light-chain framework (VL)
| [TO VERIFY]
| [TO VERIFY]
|}

The framework templates are chosen according to their sequence compatibility with the target antibody, taking into account the CDR canonical structures.

The framework is relatively conserved among antibodies, and a high sequence identity usually facilitates reliable homology modelling.

LYRA reports both sequence identity and BLOSUM62 similarity scores. These should not be confused: sequence identity measures identical residues, whereas BLOSUM62 also accounts for conservative substitutions.

=== Q2. Light-chain type ===

'''Q2: What is the light chain type? Is it the most common in humans?'''

'''Answer:'''

The SI135 antibody contains a '''lambda (λ) light chain'''.

The sequence is consistent with the IGLV3-21 light-chain family, which is associated with a well-characterized subset of CLL antibodies.

Lambda is not the most common light-chain type in humans. Approximately 60% of human antibodies contain kappa (κ) light chains and 40% contain lambda (λ) light chains, although proportions vary between individuals and tissues.

=== Q3. Sequence differences between target and template ===

'''Q3: where are the differences between target and template located? Which are the most drastic that could affect the modeling process (e.g. G->W)?'''

'''Answer:'''

The framework regions are generally more conserved than the CDRs, which are responsible for much of the sequence diversity in antibodies.

The differences observed in the SI135 alignment should be classified according to their physicochemical properties.

Examples of substitutions with potentially important structural effects include:

* G → W: introduces a large aromatic side chain in place of a very small and flexible residue.
* A → D: introduces a negatively charged side chain.
* D → K: reverses the formal side-chain charge.
* Substitutions involving proline: may restrict local backbone conformations.

Conservative substitutions, such as I → V or D → E, are generally less disruptive.

However, the structural consequences depend on the position of the substitution. A change in the buried framework can be more important than a non-conservative substitution in an exposed loop.

'''SI135-specific substitutions:''' [TO VERIFY from the LYRA alignment]

=== Q4. CDR loop lengths ===

'''Q4: Do the loops/CDR have the same length in target and template? What does it mean, in terms of expected accuracy?'''

'''Answer:'''

The CDR loop lengths must be compared with those of the selected structural templates.

The SI135 antibody contains the following CDR3 sequences:

<pre>
Heavy-chain CDR3:
ARDVNTMDV
Length: 9 amino acids

Light-chain CDR3:
QVWDSGSDHPWV
Length: 12 amino acids
</pre>

If the target and template loops have the same length and similar conformations, the model can generally be built more reliably.

Differences in loop length make modelling more difficult because insertions, deletions, or alternative loop conformations must be accommodated.

LYRA attempts to address this by selecting separate loop templates according to loop length and predicted canonical structure.

'''SI135 target/template length differences:''' [TO VERIFY]

=== Q5. Number and similarity of structural templates ===

'''Q5: How many templates were used? Which is the region for which the template is more dissimilar to the target sequence?'''

'''Answer:'''

LYRA uses structural templates for the heavy- and light-chain frameworks and can select additional templates for individual CDR loops.

There are eight modelled regions in total:

* Two framework regions (VH and VL).
* Three CDRs in the heavy chain.
* Three CDRs in the light chain.

However, the number of '''distinct PDB templates''' may be smaller than eight because several regions can be modelled using the same structure.

'''Number of distinct templates for SI135:''' [TO VERIFY]

'''Region with the least similar template:''' [TO VERIFY]

CDR-H3 is frequently a challenging region to model because of its structural diversity, although the relatively long CDR-L3 of SI135 is also of particular interest.

== Part 2 — Exploring the antibody structure ==

=== Q6. Shape of the antigen-binding site ===

'''Q6: What is the general shape of the ABS: flat, a pocket, a mountain or a groove?'''

'''Answer:'''

The antigen-binding site is formed by the six CDR loops of the variable heavy and light chains.

Its surface can be classified qualitatively as flat, pocket-like, groove-like, or protruding (mountain-like).

'''SI135 classification:''' [TO VERIFY from the LYRA structure displayed as a surface]

The classification should be based on the overall geometry of the combined CDR surface rather than the shape of a single loop.

''Teaching point:'' The shape of the antigen-binding site can provide clues about molecular recognition, but it does not directly identify the antigen.

=== Q7. Longest and most protruding CDR loop ===

'''Q7: Which loop is the longest and protruding the most? Is this the most common case?'''

'''Answer:'''

SI135 has a relatively short heavy-chain CDR3 and a long light-chain CDR3.

<pre>
CDR-H3: ARDVNTMDV (9 residues)
CDR-L3: QVWDSGSDHPWV (12 residues)
</pre>

Using the short, structurally defined CDR boundaries, '''CDR-L3 is the longest CDR in this antibody'''.

The loop that protrudes the most should be identified by inspecting the three-dimensional structure.

'''Most protruding loop:''' [TO VERIFY in PyMOL]

This is an unusual feature compared with the common situation in which CDR-H3 is the most variable and often the most prominent loop.

A long CDR does not necessarily protrude further than a short CDR, because its conformation can fold back towards the antibody framework.

== Part 3 — Somatic mutations ==

=== Q8. Mutations potentially affecting antigen binding ===

'''Q8: Which Mutations are located inside/near the CDRs? Which are the ones that are most likely to have an impact on the binding of the antibody?'''

'''Answer:'''

The somatic mutations must be identified by comparison with the closest germline V and J sequences, using IMGT/DomainGapAlign.

The SI135 sequences are consistent with IGHV3-21 and IGLV3-21 family usage, which is frequently associated with stereotyped B-cell receptors in CLL.

'''Exact somatic mutations:''' [TO VERIFY using IMGT]

Mutations within the CDRs or in framework positions supporting the CDRs are particularly interesting.

Mutations are more likely to affect binding when they:

* Change the electrostatic properties of the binding site.
* Introduce or remove aromatic residues.
* Change hydrogen-bonding possibilities.
* Alter loop conformation or flexibility.
* Affect residues exposed towards the antigen.

Mutations affecting the charge or geometry of residues near the centre of the antigen-binding site are plausible candidates for having a functional effect.

However, the impact of a mutation cannot be established confidently from the structure alone.

''Teaching point:'' Mutations in the framework may indirectly alter binding by affecting the orientation or stability of the CDR loops.

Differences introduced during V(D)J recombination, especially in CDR3, should not automatically be interpreted as somatic mutations.

== Part 4 — Electrostatic properties ==

=== Q9. Charged residues in the CDRs ===

'''Q9: Identify the residues in the CDRs that mostly contribute to the charge distribution.'''

'''Answer:'''

The most relevant charged amino acids are:

* Positively charged: arginine (R), lysine (K).
* Negatively charged: aspartate (D), glutamate (E).

For SI135, the CDR3 sequences contain several charged residues.

'''Heavy-chain CDR3:'''

<pre>
A R D V N T M D V
+ - -
</pre>

This loop contains:

* One arginine (R), positively charged.
* Two aspartates (D), negatively charged.

'''Light-chain CDR3:'''

<pre>
Q V W D S G S D H P W V
- -
</pre>

This loop contains:

* Two aspartates (D), negatively charged.
* One histidine (H), which may carry a positive charge depending on its environment and pH.

Both CDR3 loops therefore contain negatively charged residues.

Other charged residues in CDR1 and CDR2 also contribute to the overall charge distribution.

The surface distribution must be inspected in PyMOL because residues that are close in sequence are not necessarily close in three-dimensional space.

A local concentration of aspartate residues could produce a negatively charged region at the antigen-binding surface.

''Teaching point:'' Electrostatic potential reflects the spatial distribution of charges, not simply the total number of charged amino acids.

=== Q10. Effect of introducing a positive charge (optional) ===

'''Q10: (optional) Mutate some residues in order to have a positive charge in the middle of the antigen binding site, then recalculate the electrostatic potentials. Can you see the changes?'''

'''Answer:'''

Introducing a positively charged residue, such as lysine (K) or arginine (R), should increase the local positive electrostatic potential.

For example, replacing an exposed aspartate (D) with lysine (K) in a CDR could reverse the local charge.

After recalculating the electrostatic potential, a change in surface coloring should become visible near the mutated residue.

The extent of the change depends on the location and conformation of the new side chain.

A more positive surface does not necessarily imply stronger antigen binding, because the properties of the antigen must also be considered.

== Part 5 — Comparison of modelling approaches ==

=== Q11. Comparison between LYRA and ABodyBuilder2 ===

'''Q11: Go to the AbodyBuilder 2 site and make a new structure prediction. Download the PDB file and compare with the model generated by LYRA. Can you see some differences?'''

'''Answer:'''

LYRA and ABodyBuilder2 use different approaches to antibody structure prediction.

{| class="wikitable"
! LYRA
! ABodyBuilder2
|-
| Traditional template-based homology modelling
| Deep-learning-based structure prediction
|-
| Selects framework and CDR templates from known experimental structures
| Uses a neural network trained on antibody structures
|-
| Uses canonical loop structures and loop grafting
| Learns structural patterns from training data
|-
| Quality depends strongly on suitable templates
| Provides prediction-error estimates for different antibody regions
|}

The overall immunoglobulin fold is expected to be similar in both models.

Differences may occur in:

* CDR loop conformations, particularly CDR-H3 or unusual loops.
* Relative orientations of the heavy and light chains.
* Side-chain conformations.

'''Specific differences between the SI135 models:''' [TO VERIFY after structural superposition]

The models can be compared in PyMOL using:

<pre>
align ABodyBuilder2, LYRA
</pre>

Replace the object names with those used in the PyMOL session.

The RMSD obtained after structural alignment provides a quantitative measure of structural differences.

''Teaching point:'' Agreement between two predictions can increase confidence in a structural hypothesis, but does not establish experimental accuracy.

== Summary ==

This exercise illustrates several important principles:

* Antibody frameworks are relatively conserved, whereas CDR loops provide sequence and structural diversity.
* Template-based modelling is influenced by sequence similarity, loop length, and canonical structure.
* CDR-H3 is often particularly variable, but other CDRs may also present modelling challenges.
* Somatic mutations may affect antigen recognition by changing the structural or chemical properties of the binding surface.
* Electrostatic properties depend on the three-dimensional arrangement of charged residues.
* Deep-learning tools provide an alternative to traditional homology modelling, but predicted structures must still be evaluated critically.

== References ==

* [https://services.healthtech.dtu.dk/cgi-bin/webface2.cgi?jobid=6AD8166000282F67166ED625&wait=20 SI135 LYRA modelling results]
* [https://services.healthtech.dtu.dk/services/LYRA-1.0/ LYRA — DTU Health Tech]
* [https://www.imgt.org/ IMGT immunogenetics]
* [https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/abodybuilder2/ ABodyBuilder2]
* [https://pmc.ncbi.nlm.nih.gov/articles/PMC7015129/ Antibody sequences and antigen recognition in CLL]

Exercise:AntibodyStructurePrediction ans

2026-10-21T02:18:32Z

Carol: Created page with " __TOC__ = Antibody Structure Modelling — Teacher's Answers = '''Case study:''' SI135 antibody from chronic lymphocytic leukemia (CLL) '''LYRA prediction:''' [https://services.healthtech.dtu.dk/cgi-bin/webface2.cgi?jobid=6AD8166000282F67166ED625&wait=20 View the original modelling job] ''This page provides expected answers to Q1–Q11. Values marked [TO VERIFY] require inspection of the LYRA model, sequence alignment, or additional analyses.'' == Part 1 — Antibo..."

__TOC__

= Antibody Structure Modelling — Teacher's Answers =

'''Case study:''' SI135 antibody from chronic lymphocytic leukemia (CLL)

'''LYRA prediction:''' [https://services.healthtech.dtu.dk/cgi-bin/webface2.cgi?jobid=6AD8166000282F67166ED625&wait=20 View the original modelling job]

''This page provides expected answers to Q1–Q11. Values marked [TO VERIFY] require inspection of the LYRA model, sequence alignment, or additional analyses.''

== Part 1 — Antibody modelling with LYRA ==

=== Q1. Framework template sequence similarity ===

'''Q1: what is the sequence similarity of the best H and L templates for the framework (FW)?'''

'''Answer:'''

{| class="wikitable"
! Region
! Selected template
! Sequence identity
|-
| Heavy-chain framework (VH)
| [TO VERIFY]
| [TO VERIFY]
|-
| Light-chain framework (VL)
| [TO VERIFY]
| [TO VERIFY]
|}

The framework templates are chosen according to their sequence compatibility with the target antibody, taking into account the CDR canonical structures.

The framework is relatively conserved among antibodies, and a high sequence identity usually facilitates reliable homology modelling.

LYRA reports both sequence identity and BLOSUM62 similarity scores. These should not be confused: sequence identity measures identical residues, whereas BLOSUM62 also accounts for conservative substitutions.

=== Q2. Light-chain type ===

'''Q2: What is the light chain type? Is it the most common in humans?'''

'''Answer:'''

The SI135 antibody contains a '''lambda (λ) light chain'''.

The sequence is consistent with the IGLV3-21 light-chain family, which is associated with a well-characterized subset of CLL antibodies.

Lambda is not the most common light-chain type in humans. Approximately 60% of human antibodies contain kappa (κ) light chains and 40% contain lambda (λ) light chains, although proportions vary between individuals and tissues.

=== Q3. Sequence differences between target and template ===

'''Q3: where are the differences between target and template located? Which are the most drastic that could affect the modeling process (e.g. G->W)?'''

'''Answer:'''

The framework regions are generally more conserved than the CDRs, which are responsible for much of the sequence diversity in antibodies.

The differences observed in the SI135 alignment should be classified according to their physicochemical properties.

Examples of substitutions with potentially important structural effects include:

* G → W: introduces a large aromatic side chain in place of a very small and flexible residue.
* A → D: introduces a negatively charged side chain.
* D → K: reverses the formal side-chain charge.
* Substitutions involving proline: may restrict local backbone conformations.

Conservative substitutions, such as I → V or D → E, are generally less disruptive.

However, the structural consequences depend on the position of the substitution. A change in the buried framework can be more important than a non-conservative substitution in an exposed loop.

'''SI135-specific substitutions:''' [TO VERIFY from the LYRA alignment]

=== Q4. CDR loop lengths ===

'''Q4: Do the loops/CDR have the same length in target and template? What does it mean, in terms of expected accuracy?'''

'''Answer:'''

The CDR loop lengths must be compared with those of the selected structural templates.

The SI135 antibody contains the following CDR3 sequences:

<pre>
Heavy-chain CDR3:
ARDVNTMDV
Length: 9 amino acids

Light-chain CDR3:
QVWDSGSDHPWV
Length: 12 amino acids
</pre>

If the target and template loops have the same length and similar conformations, the model can generally be built more reliably.

Differences in loop length make modelling more difficult because insertions, deletions, or alternative loop conformations must be accommodated.

LYRA attempts to address this by selecting separate loop templates according to loop length and predicted canonical structure.

'''SI135 target/template length differences:''' [TO VERIFY]

=== Q5. Number and similarity of structural templates ===

'''Q5: How many templates were used? Which is the region for which the template is more dissimilar to the target sequence?'''

'''Answer:'''

LYRA uses structural templates for the heavy- and light-chain frameworks and can select additional templates for individual CDR loops.

There are eight modelled regions in total:

* Two framework regions (VH and VL).
* Three CDRs in the heavy chain.
* Three CDRs in the light chain.

However, the number of '''distinct PDB templates''' may be smaller than eight because several regions can be modelled using the same structure.

'''Number of distinct templates for SI135:''' [TO VERIFY]

'''Region with the least similar template:''' [TO VERIFY]

CDR-H3 is frequently a challenging region to model because of its structural diversity, although the relatively long CDR-L3 of SI135 is also of particular interest.

== Part 2 — Exploring the antibody structure ==

=== Q6. Shape of the antigen-binding site ===

'''Q6: What is the general shape of the ABS: flat, a pocket, a mountain or a groove?'''

'''Answer:'''

The antigen-binding site is formed by the six CDR loops of the variable heavy and light chains.

Its surface can be classified qualitatively as flat, pocket-like, groove-like, or protruding (mountain-like).

'''SI135 classification:''' [TO VERIFY from the LYRA structure displayed as a surface]

The classification should be based on the overall geometry of the combined CDR surface rather than the shape of a single loop.

''Teaching point:'' The shape of the antigen-binding site can provide clues about molecular recognition, but it does not directly identify the antigen.

=== Q7. Longest and most protruding CDR loop ===

'''Q7: Which loop is the longest and protruding the most? Is this the most common case?'''

'''Answer:'''

SI135 has a relatively short heavy-chain CDR3 and a long light-chain CDR3.

<pre>
CDR-H3: ARDVNTMDV (9 residues)
CDR-L3: QVWDSGSDHPWV (12 residues)
</pre>

Using the short, structurally defined CDR boundaries, '''CDR-L3 is the longest CDR in this antibody'''.

The loop that protrudes the most should be identified by inspecting the three-dimensional structure.

'''Most protruding loop:''' [TO VERIFY in PyMOL]

This is an unusual feature compared with the common situation in which CDR-H3 is the most variable and often the most prominent loop.

A long CDR does not necessarily protrude further than a short CDR, because its conformation can fold back towards the antibody framework.

== Part 3 — Somatic mutations ==

=== Q8. Mutations potentially affecting antigen binding ===

'''Q8: Which Mutations are located inside/near the CDRs? Which are the ones that are most likely to have an impact on the binding of the antibody?'''

'''Answer:'''

The somatic mutations must be identified by comparison with the closest germline V and J sequences, using IMGT/DomainGapAlign.

The SI135 sequences are consistent with IGHV3-21 and IGLV3-21 family usage, which is frequently associated with stereotyped B-cell receptors in CLL.

'''Exact somatic mutations:''' [TO VERIFY using IMGT]

Mutations within the CDRs or in framework positions supporting the CDRs are particularly interesting.

Mutations are more likely to affect binding when they:

* Change the electrostatic properties of the binding site.
* Introduce or remove aromatic residues.
* Change hydrogen-bonding possibilities.
* Alter loop conformation or flexibility.
* Affect residues exposed towards the antigen.

Mutations affecting the charge or geometry of residues near the centre of the antigen-binding site are plausible candidates for having a functional effect.

However, the impact of a mutation cannot be established confidently from the structure alone.

''Teaching point:'' Mutations in the framework may indirectly alter binding by affecting the orientation or stability of the CDR loops.

Differences introduced during V(D)J recombination, especially in CDR3, should not automatically be interpreted as somatic mutations.

== Part 4 — Electrostatic properties ==

=== Q9. Charged residues in the CDRs ===

'''Q9: Identify the residues in the CDRs that mostly contribute to the charge distribution.'''

'''Answer:'''

The most relevant charged amino acids are:

* Positively charged: arginine (R), lysine (K).
* Negatively charged: aspartate (D), glutamate (E).

For SI135, the CDR3 sequences contain several charged residues.

'''Heavy-chain CDR3:'''

<pre>
A R D V N T M D V
+ - -
</pre>

This loop contains:

* One arginine (R), positively charged.
* Two aspartates (D), negatively charged.

'''Light-chain CDR3:'''

<pre>
Q V W D S G S D H P W V
- -
</pre>

This loop contains:

* Two aspartates (D), negatively charged.
* One histidine (H), which may carry a positive charge depending on its environment and pH.

Both CDR3 loops therefore contain negatively charged residues.

Other charged residues in CDR1 and CDR2 also contribute to the overall charge distribution.

The surface distribution must be inspected in PyMOL because residues that are close in sequence are not necessarily close in three-dimensional space.

A local concentration of aspartate residues could produce a negatively charged region at the antigen-binding surface.

''Teaching point:'' Electrostatic potential reflects the spatial distribution of charges, not simply the total number of charged amino acids.

=== Q10. Effect of introducing a positive charge (optional) ===

'''Q10: (optional) Mutate some residues in order to have a positive charge in the middle of the antigen binding site, then recalculate the electrostatic potentials. Can you see the changes?'''

'''Answer:'''

Introducing a positively charged residue, such as lysine (K) or arginine (R), should increase the local positive electrostatic potential.

For example, replacing an exposed aspartate (D) with lysine (K) in a CDR could reverse the local charge.

After recalculating the electrostatic potential, a change in surface coloring should become visible near the mutated residue.

The extent of the change depends on the location and conformation of the new side chain.

A more positive surface does not necessarily imply stronger antigen binding, because the properties of the antigen must also be considered.

== Part 5 — Comparison of modelling approaches ==

=== Q11. Comparison between LYRA and ABodyBuilder2 ===

'''Q11: Go to the AbodyBuilder 2 site and make a new structure prediction. Download the PDB file and compare with the model generated by LYRA. Can you see some differences?'''

'''Answer:'''

LYRA and ABodyBuilder2 use different approaches to antibody structure prediction.

{| class="wikitable"
! LYRA
! ABodyBuilder2
|-
| Traditional template-based homology modelling
| Deep-learning-based structure prediction
|-
| Selects framework and CDR templates from known experimental structures
| Uses a neural network trained on antibody structures
|-
| Uses canonical loop structures and loop grafting
| Learns structural patterns from training data
|-
| Quality depends strongly on suitable templates
| Provides prediction-error estimates for different antibody regions
|}

The overall immunoglobulin fold is expected to be similar in both models.

Differences may occur in:

* CDR loop conformations, particularly CDR-H3 or unusual loops.
* Relative orientations of the heavy and light chains.
* Side-chain conformations.

'''Specific differences between the SI135 models:''' [TO VERIFY after structural superposition]

The models can be compared in PyMOL using:

<pre>
align ABodyBuilder2, LYRA
</pre>

Replace the object names with those used in the PyMOL session.

The RMSD obtained after structural alignment provides a quantitative measure of structural differences.

''Teaching point:'' Agreement between two predictions can increase confidence in a structural hypothesis, but does not establish experimental accuracy.

== Summary ==

This exercise illustrates several important principles:

* Antibody frameworks are relatively conserved, whereas CDR loops provide sequence and structural diversity.
* Template-based modelling is influenced by sequence similarity, loop length, and canonical structure.
* CDR-H3 is often particularly variable, but other CDRs may also present modelling challenges.
* Somatic mutations may affect antigen recognition by changing the structural or chemical properties of the binding surface.
* Electrostatic properties depend on the three-dimensional arrangement of charged residues.
* Deep-learning tools provide an alternative to traditional homology modelling, but predicted structures must still be evaluated critically.

== References ==

* [https://services.healthtech.dtu.dk/cgi-bin/webface2.cgi?jobid=6AD8166000282F67166ED625&wait=20 SI135 LYRA modelling results]
* [https://services.healthtech.dtu.dk/services/LYRA-1.0/ LYRA — DTU Health Tech]
* [https://www.imgt.org/ IMGT immunogenetics]
* [https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/abodybuilder2/ ABodyBuilder2]
* [https://pmc.ncbi.nlm.nih.gov/articles/PMC7015129/ Antibody sequences and antigen recognition in CLL]

Exercise:AntibodyStructurePrediction

2026-10-21T02:14:42Z

Carol: /* Part 1 — Identify the antibody regions */

= Antibody Structure Modelling: From Sequence to 3D Structure =

__TOC__

== Introduction: ==

Antibodies recognize antigens through their variable domains, which contain six complementarity-determining regions (CDRs): three in the heavy chain and three in the light chain. These loops form the antigen-binding site and are supported by more conserved framework (FW) regions.

Predicting the three-dimensional structure of antibodies helps us understand how their sequence, shape, and chemical properties may influence antigen recognition. Different computational approaches, from traditional template-based modelling to modern deep-learning methods, can be used to predict antibody structures.

In this exercise, you will model an antibody from a patient with chronic lymphocytic leukemia (CLL), explore its structure using PyMOL, and investigate how sequence variation and structural features may influence antigen binding.

== Learning objectives ==

By the end of this exercise, you should be able to:

* Identify the framework and CDR regions of an antibody.
* Understand the basic principles of template-based antibody modelling.
* Generate and visualize a three-dimensional antibody model using computational tools.
* Explore how CDR sequence, structure, and electrostatic properties may influence antigen recognition.
* Identify somatic mutations and discuss their potential effects on antibody binding.
* Compare antibody structures predicted using traditional template-based and modern deep-learning approaches.

== Antibody sequences ==

We will use the variable regions of the SI135 antibody.

'''Light chain (VL):'''

<pre>
>SI135_VL
SYVVTQPPSVSVAPGKTARITCGGNNIGSKSVHWYQQKPGQAPVLVIYYDNDRPSGIPERFSGSNSGNTATLTISRVEAGDEADYYCQVWDSGSDHPWVFGGGTKLTVL
</pre>

'''Heavy chain (VH):'''

<pre>
>SI135_VH
EVQLVESGGGLVRPGGSLRLSCAASGFTFSTYSMNWVRQAPGKGLEWVSSISSSSNSIYCADSVKGRFTISRDNANNSLYMQMNSLRAEDTAVYYCARDVNTMDVWGQGILVTVLL
</pre>

== Part 1 — Identify the antibody regions ==

Go to the [https://services.healthtech.dtu.dk/services/LYRA-1.0/ Lyra] web page.

''Lyra is a tool for predicting antibody and TCR structures. It is possible to use Lyra in automated or expert mode.''

Insert the light and heavy chains in the proper text area and submit. After a few seconds you will arrive to the output page, where the alignment and all the templates are displayed.

Note that for each Region of each chain of the antibody a Template has been found.

{| class="wikitable"
! Region
! Templates
|-
| Framework
| ...
|-
| Loop 1 (CDR1)
| ...
|-
| Loop 2 (CDR2)
| ...
|-
| Loop 3 (CDR3)
| ...
|}

Note that after the Regions there is an extra template used for packing the whole structure.

Explore the output page and the different tabs - check the PDB file, Log, and manual selection ones to find this info.

Now answer the following questions:

'''Q1: what is the sequence similarity of the best H and L templates for the framework (FW)?'''

'''Q2: What is the light chain type? Is it the most common in humans?'''

=== Sequence alignment and CDR modelling ===

Focus on the sequence alignment used to generate the model. You can find it in one of the tabs in Lyra website, as well as in the PDB file itself.

'''Q3: where are the differences between target and template located? Which are the most drastic that could affect the modeling process (e.g. G->W)?'''

''hint: if you don't remeber the amino acids, you can use the Blosum62 matrix to estimate the expected effect of a substitution: the lower the number, the larger the expected impact. For sake of time, you only need to give a qualitative answer.''

'''Q4: Do the loops/CDR have the same length in target and template? What does it mean, in terms of expected accuracy?'''

Now, look at the templates automatically selected by Lyra to build the final model.

'''Q5: How many templates were used? Which is the region for which the template is more dissimilar to the target sequence?'''

Download the pdb file and open it with pymol. There is a small "S" bottom right, it shows/hides the sequence. Show the whole protein as cartoon and color by chain. Select and color in red all the CDRs. (the selected residues appear as the "sele" item in the menu). The CDR definition can be seen in the alignment generated by Lyra.

'''Q6: What is the general shape of the ABS: flat, a pocket, a mountain or a groove?''' ''It can help to show the antibody as surface''

'''Q7: Which loop is the longest and protruding the most? Is this the most common case?'''

Use the [http://www.imgt.org/3Dstructure-DB/cgi/DomainGapAlign.cgi Domain Gap Align tool] identify the somatic mutations in the V and J regions. You need to use the amino acid sequence and select human as the source organism. In PyMol, show the mutations as sticks (main chain and side chain) and color them green.

'''Q8: Which Mutations are located inside/near the CDRs? Which are the ones that are most likely to have an impact on the binding of the antibody?''' ''A qualitative answer is sufficient''

Generate the electrostatic potentials of the Ig ('''A'''ction -> generate -> vacuum electrostatic -> protein contact potentials).

'''Q9: Identify the residues in the CDRs that mostly contribute to the charge distribution.'''

''highlight them in the sequence and look for charged residues such as K, R, D, E''

'''Q10: (optional) Mutate some residues in order to have a positive charge in the middle of the antigen binding site, then recalculate the electrostatic potentials. Can you see the changes?'''

'''Q11: Go to the AbodyBuilder 2 site and make a new structure prediction. Download the PDB file and compare with the model generated by LYRA. Can you see some differences?'''

'''Finished!'''

Exercise:AntibodyStructurePrediction

2026-10-21T02:14:20Z

Carol:

= Antibody Structure Modelling: From Sequence to 3D Structure =

__TOC__

== Introduction: ==

Antibodies recognize antigens through their variable domains, which contain six complementarity-determining regions (CDRs): three in the heavy chain and three in the light chain. These loops form the antigen-binding site and are supported by more conserved framework (FW) regions.

Predicting the three-dimensional structure of antibodies helps us understand how their sequence, shape, and chemical properties may influence antigen recognition. Different computational approaches, from traditional template-based modelling to modern deep-learning methods, can be used to predict antibody structures.

In this exercise, you will model an antibody from a patient with chronic lymphocytic leukemia (CLL), explore its structure using PyMOL, and investigate how sequence variation and structural features may influence antigen binding.

== Learning objectives ==

By the end of this exercise, you should be able to:

* Identify the framework and CDR regions of an antibody.
* Understand the basic principles of template-based antibody modelling.
* Generate and visualize a three-dimensional antibody model using computational tools.
* Explore how CDR sequence, structure, and electrostatic properties may influence antigen recognition.
* Identify somatic mutations and discuss their potential effects on antibody binding.
* Compare antibody structures predicted using traditional template-based and modern deep-learning approaches.

== Antibody sequences ==

We will use the variable regions of the SI135 antibody.

'''Light chain (VL):'''

<pre>
>SI135_VL
SYVVTQPPSVSVAPGKTARITCGGNNIGSKSVHWYQQKPGQAPVLVIYYDNDRPSGIPERFSGSNSGNTATLTISRVEAGDEADYYCQVWDSGSDHPWVFGGGTKLTVL
</pre>

'''Heavy chain (VH):'''

<pre>
>SI135_VH
EVQLVESGGGLVRPGGSLRLSCAASGFTFSTYSMNWVRQAPGKGLEWVSSISSSSNSIYCADSVKGRFTISRDNANNSLYMQMNSLRAEDTAVYYCARDVNTMDVWGQGILVTVLL
</pre>

== Part 1 — Identify the antibody regions ==

Go to the [https://services.healthtech.dtu.dk/services/LYRA-1.0/ Lyra] web page.

''Lyra is a tool for predicting antibody and TCR structures. It is possible to use Lyra in automated or expert mode.''

Insert the light and heavy chains in the proper text area and submit. After a few seconds you will arrive to the output page, where the alignment and all the templates are displayed.

Note that for each Region of each chain of the antibody a Template has been found.

{| class="wikitable"
! Region
! Templates
|-
| Framework
| ...
|-
| Loop 1 (CDR1)
| ...
|-
| Loop 2 (CDR2)
| ...
|-
| Loop 3 (CDR3)
| ...
|}

Note that after the Regions there is an extra template used for packing the whole structure.

Explore the output page and the different tabs - check the PDB file, Log, and manual selection ones to find this info.

Now answer the following questions:

'''Q1: what is the sequence similarity of the best H and L templates for the framework (FW)?'''

'''Q2: What is the light chain type? Is it the most common in humans?'''

=== Sequence alignment and CDR modelling ===

Focus on the sequence alignment used to generate the model. You can find it in one of the tabs in Lyra website, as well as in the PDB file itself.

'''Q3: where are the differences between target and template located? Which are the most drastic that could affect the modeling process (e.g. G->W)?'''

''hint: if you don't remeber the amino acids, you can use the Blosum62 matrix to estimate the expected effect of a substitution: the lower the number, the larger the expected impact. For sake of time, you only need to give a qualitative answer.''

'''Q4: Do the loops/CDR have the same length in target and template? What does it mean, in terms of expected accuracy?'''

Now, look at the templates automatically selected by Lyra to build the final model.

'''Q5: How many templates were used? Which is the region for which the template is more dissimilar to the target sequence?'''

Download the pdb file and open it with pymol. There is a small "S" bottom right, it shows/hides the sequence. Show the whole protein as cartoon and color by chain. Select and color in red all the CDRs. (the selected residues appear as the "sele" item in the menu). The CDR definition can be seen in the alignment generated by Lyra.

'''Q6: What is the general shape of the ABS: flat, a pocket, a mountain or a groove?''' ''It can help to show the antibody as surface''

'''Q7: Which loop is the longest and protruding the most? Is this the most common case?'''

Use the [http://www.imgt.org/3Dstructure-DB/cgi/DomainGapAlign.cgi Domain Gap Align tool] identify the somatic mutations in the V and J regions. You need to use the amino acid sequence and select human as the source organism. In PyMol, show the mutations as sticks (main chain and side chain) and color them green.

'''Q8: Which Mutations are located inside/near the CDRs? Which are the ones that are most likely to have an impact on the binding of the antibody?''' ''A qualitative answer is sufficient''

Generate the electrostatic potentials of the Ig ('''A'''ction -> generate -> vacuum electrostatic -> protein contact potentials).

'''Q9: Identify the residues in the CDRs that mostly contribute to the charge distribution.'''

''highlight them in the sequence and look for charged residues such as K, R, D, E''

'''Q10: (optional) Mutate some residues in order to have a positive charge in the middle of the antigen binding site, then recalculate the electrostatic potentials. Can you see the changes?'''

'''Q11: Go to the AbodyBuilder 2 site and make a new structure prediction. Download the PDB file and compare with the model generated by LYRA. Can you see some differences?'''

'''Finished!'''

Exercise:AntibodyStructurePrediction

2026-10-21T02:08:44Z

Carol: /* Part 1 — Identify the antibody regions */

= Antibody Structure Modelling: From Sequence to 3D Structure =

__TOC__

== Introduction: ==

Antibodies recognize antigens through their variable domains, which contain six complementarity-determining regions (CDRs): three in the heavy chain and three in the light chain. These loops form the antigen-binding site and are supported by more conserved framework (FW) regions.

Predicting the three-dimensional structure of antibodies helps us understand how their sequence, shape, and chemical properties may influence antigen recognition. Different computational approaches, from traditional template-based modelling to modern deep-learning methods, can be used to predict antibody structures.

In this exercise, you will model an antibody from a patient with chronic lymphocytic leukemia (CLL), explore its structure using PyMOL, and investigate how sequence variation and structural features may influence antigen binding.

== Learning objectives ==

By the end of this exercise, you should be able to:

* Identify the framework and CDR regions of an antibody.
* Understand the basic principles of template-based antibody modelling.
* Generate and visualize a three-dimensional antibody model using computational tools.
* Explore how CDR sequence, structure, and electrostatic properties may influence antigen recognition.
* Identify somatic mutations and discuss their potential effects on antibody binding.
* Compare antibody structures predicted using traditional template-based and modern deep-learning approaches.

== Antibody sequences ==

We will use the variable regions of the SI135 antibody.

'''Light chain (VL):'''

<pre>
>SI135_VL
SYVVTQPPSVSVAPGKTARITCGGNNIGSKSVHWYQQKPGQAPVLVIYYDNDRPSGIPERFSGSNSGNTATLTISRVEAGDEADYYCQVWDSGSDHPWVFGGGTKLTVL
</pre>

'''Heavy chain (VH):'''

<pre>
>SI135_VH
EVQLVESGGGLVRPGGSLRLSCAASGFTFSTYSMNWVRQAPGKGLEWVSSISSSSNSIYCADSVKGRFTISRDNANNSLYMQMNSLRAEDTAVYYCARDVNTMDVWGQGILVTVLL
</pre>

== Part 1 — Identify the antibody regions ==

Go to the [https://services.healthtech.dtu.dk/services/LYRA-1.0/ Lyra] web page.

''Lyra is a tool for predicting antibody and TCR structures. It is possible to use Lyra in automated or expert mode.''

Insert the light and heavy chains in the proper text area and submit. After a few seconds you will arrive to the output page, where the alignment and all the templates are displayed.

Note that for each Region of each chain of the antibody a Template has been found.

{| class="wikitable"
! Region
! Templates
|-
| Framework
| ...
|-
| Loop 1 (CDR1)
| ...
|-
| Loop 2 (CDR2)
| ...
|-
| Loop 3 (CDR3)
| ...
|}

Note that after the Regions there is an extra template used for packing the whole structure.

Explore the output page and the different tabs - check the PDB file, Log, and manual selection ones to find this info.

Now answer the following questions:

'''Q1: what is the sequence similarity of the best H and L templates for the framework (FW)?'''

'''Q2: What is the light chain type? Is it the most common in humans?'''

Focus on the sequence alignment used to generate the model. You can find it in one of the tabs in Lyra website, as well as in the PDB file itself.

'''Q3: where are the differences between target and template located? Which are the most drastic that could affect the modeling process (e.g. G->W)?'''

''hint: if you don't remeber the amino acids, you can use the Blosum62 matrix to estimate the expected effect of a substitution: the lower the number, the larger the expected impact. For sake of time, you only need to give a qualitative answer.''

'''Q4: Do the loops/CDR have the same length in target and template? What does it mean, in terms of expected accuracy?'''

Now, look at the templates automatically selected by Lyra to build the final model.

'''Q5: How many templates were used? Which is the region for which the template is more dissimilar to the target sequence?'''

Download the pdb file and open it with pymol. There is a small "S" bottom right, it shows/hides the sequence. Show the whole protein as cartoon and color by chain. Select and color in red all the CDRs. (the selected residues appear as the "sele" item in the menu). The CDR definition can be seen in the alignment generated by Lyra.

'''Q6: What is the general shape of the ABS: flat, a pocket, a mountain or a groove?''' ''It can help to show the antibody as surface''

'''Q7: Which loop is the longest and protruding the most? Is this the most common case?'''

Use the [http://www.imgt.org/3Dstructure-DB/cgi/DomainGapAlign.cgi Domain Gap Align tool] identify the somatic mutations in the V and J regions. You need to use the amino acid sequence and select human as the source organism. In PyMol, show the mutations as sticks (main chain and side chain) and color them green.

'''Q8: Which Mutations are located inside/near the CDRs? Which are the ones that are most likely to have an impact on the binding of the antibody?''' ''A qualitative answer is sufficient''

Generate the electrostatic potentials of the Ig ('''A'''ction -> generate -> vacuum electrostatic -> protein contact potentials).

'''Q9: Identify the residues in the CDRs that mostly contribute to the charge distribution.'''

''highlight them in the sequence and look for charged residues such as K, R, D, E''

'''Q10: (optional) Mutate some residues in order to have a positive charge in the middle of the antigen binding site, then recalculate the electrostatic potentials. Can you see the changes?'''

'''Q11: Go to the AbodyBuilder 2 site and make a new structure prediction. Download the PDB file and compare with the model generated by LYRA. Can you see some differences?'''

'''Finished!'''