Antibody-Antigen Structure Prediction for Therapeutic Development

Antibody-antigen structure prediction is revolutionizing therapeutic development. Learn how to predict antibody-antigen complexes, map epitopes, and guide affinity maturation with AlphaFold2-Multimer.

#Why Antibody Structure Prediction Matters

Therapeutic antibodies represent a $150B+ market. Understanding antibody-antigen interactions accelerates:

• Epitope mapping for vaccine design
• Affinity maturation and humanization
• Developability assessment
• Biosimilar development

#Antibody Structure Basics

Antibody Architecture

Standard IgG antibody consists of:

• Heavy chain (HC): ~450 residues, VH-CH1-CH2-CH3 domains
• Light chain (LC): ~220 residues, VL-CL domains
• CDRs: 6 hypervariable loops (3 on HC, 3 on LC)
• Framework regions: Conserved β-sheet scaffold

Complementarity-Determining Regions

#Prediction Workflow

Step 1: Sequence Preparation

Prepare multi-chain FASTA format:

>heavy_chain
EVQLVESGGGLVQPGGSLRLSCAASGFTFSSYAMSWVRQAPGKGLEWVSSISGSSGYTYYAYWGKGTLVTVSS...

>light_chain
DIQMTQSPSSLSASVGDRVTITCRASQSISSYLNWYQQKPGKAPKLLIYAASSLQSGVPSRFSGSGSGTDFTL...

>antigen
MKFLKFSLLTAVLLSVVFAFSSCGDDDDTGYLPPSQAIQDLLKRMKVRFLYSNGQPPEKPEGCQNPDCFQPPQ...

Step 2: Running AlphaFold2-Multimer

Key settings for antibody predictions:

• Use AlphaFold2-Multimer (not standard AlphaFold2)
• Include both heavy and light chains
• Run all 5 models for ensemble analysis
• Check ipTM for interface confidence

Step 3: Result Validation

#Epitope Mapping

Identifying the Epitope

From the predicted complex, extract interface residues:

from Bio.PDB import PDBParser, NeighborSearch

parser = PDBParser()
structure = parser.get_structure('complex', 'antibody_antigen.pdb')

# Get antibody and antigen atoms
ab_atoms = [atom for atom in structure.get_atoms()
            if atom.get_parent().get_parent().id in ['H', 'L']]
ag_atoms = [atom for atom in structure.get_atoms()
            if atom.get_parent().get_parent().id == 'A']

# Find interface (< 5Å distance)
ns = NeighborSearch(ab_atoms)
epitope_residues = set()

for ag_atom in ag_atoms:
    neighbors = ns.search(ag_atom.coord, 5.0, level='R')
    if neighbors:
        epitope_residues.add(ag_atom.get_parent())

print(f"Epitope residues: {[r.id[1] for r in epitope_residues]}")

Paratope Analysis

Analyze which CDRs contribute to binding:

• CDR-H3: Usually dominates binding (longest and most variable)
• CDR-H1/H2: Frame the binding site
• CDR-L1/L2/L3: Complement heavy chain CDRs

#Guiding Affinity Maturation

Identifying Hotspot Residues

Use the structure to identify mutation candidates:

• Residues at interface with suboptimal contacts
• Positions with space for larger side chains
• Regions with poor shape complementarity
• Areas with unfavorable electrostatics

Computational Mutagenesis

# Example: Test mutations computationally
import pyrosetta
pyrosetta.init()

pose = pyrosetta.pose_from_pdb('complex.pdb')

# Create score function
scorefxn = pyrosetta.get_fa_scorefxn()

# Test mutation H-CDR3 position 100 to Tryptophan
from pyrosetta.toolbox import mutants
mutant = mutants.mutate_residue(pose, 100, 'W')

# Score and compare
original_score = scorefxn(pose)
mutant_score = scorefxn(mutant)

print(f"ΔΔG = {mutant_score - original_score:.2f} REU")

Antibody Library Design

#Developability Assessment

Aggregation Risk

Check for developability liabilities:

• Hydrophobic patches: > 600 Ų exposed hydrophobic surface
• Charge patches: Clustered positive or negative charges
• Unpaired cysteines: Potential for misfolding
• PTM sites: Deamidation (N-G), oxidation (M, W)

Humanization Strategies

For therapeutic development from mouse antibodies:

• Identify framework regions for humanization
• Preserve CDR structures (don't mutate supporting residues)
• Use structure to predict which backmutations needed
• Validate with AlphaFold2 after humanization

#Special Cases

Nanobodies (VHH)

Single-domain antibodies from camelids:

• Simpler input (single chain + antigen)
• Longer CDR-H3 (up to 25 residues)
• Can access cryptic epitopes
• Excellent AlphaFold2 accuracy for nanobodies

Bispecific Antibodies

# Input for bispecific (2 antigens)
>heavy_chain_1
EVQLVESGGGLVQPGG...
>light_chain_1
DIQMTQSPSSLSASVG...
>heavy_chain_2
QVQLVQSGAEVKKPGA...
>light_chain_2
DIVMTQSPDSLAVSLG...
>antigen_1
MKTFLILTTLLTQVSS...
>antigen_2
MAPAMEKKFEAAAAMQ...

#Experimental Validation

Validation Approaches

• Surface Plasmon Resonance (SPR): Measure binding kinetics
• Biolayer Interferometry (BLI): Rapid affinity measurements
• Epitope mapping: HDX-MS, mutagenesis, peptide arrays
• Structural: X-ray crystallography, cryo-EM

Functional Assays

• Neutralization assays (for viral antigens)
• Cell-based binding assays
• Competition ELISAs
• Epitope binning

#Case Studies

Case 1: COVID-19 Neutralizing Antibody

Target: SARS-CoV-2 Spike RBD
Antibody: Therapeutic mAb candidate

Results:
- ipTM: 0.78 (high confidence)
- Epitope: Overlaps ACE2 binding site
- Key contacts: CDR-H3 Y102, CDR-L1 N32
- Predicted KD: Low nM range

Validation:
- SPR confirmed nM affinity
- Neutralization IC50: 15 ng/mL
- Cryo-EM structure RMSD: 1.2 Å vs prediction

Case 2: Affinity Maturation

Starting antibody: KD = 50 nM
AlphaFold2 analysis identified:
- Suboptimal CDR-H3 contacts
- Cavity allowing for larger residue

Mutations tested:
- H-CDR3 S100W: Predicted ΔΔG = -1.8 kcal/mol
- H-CDR3 S100Y: Predicted ΔΔG = -1.2 kcal/mol

Experimental results:
- S100W variant: KD = 8 nM (6-fold improvement)
- S100Y variant: KD = 15 nM (3-fold improvement)

#Tools and Resources

• ABlooper: CDR loop structure prediction
• ANARCI: Antibody numbering and alignment
• SAbDab: Structural antibody database
• Parapred: Paratope prediction
• Rosetta: Antibody design and modeling

#Best Practices Summary