Exercise:AntibodyStructurePrediction ans

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:

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:

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.

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:

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:

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:

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.

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:

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:

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

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.

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:

align ABodyBuilder2, LYRA

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

• LYRA
• IMGT
• ABodyBuilder2