Exercise:AntibodyStructurePrediction ans

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:

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:

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.

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:

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):

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:

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.

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.