HAIRpred2 identifies antibody-interacting residues in human antigens using 3D structural features - RSA, physicochemical properties, and a pre-trained Random Forest model achieving AUC 0.78.
Unlike sequence-based tools, HAIRpred2 uses 3D structural information for residue-level epitope prediction.
Accepts standard PDB files. Uses DSSP to compute Relative Solvent Accessibility (RSA) for each residue directly from 3D coordinates.
Trained exclusively on 277 human Ag-Ab complexes from SAbDab — captures unique features of human immune recognition.
15-residue sliding window encodes RSA + pI, pKa, hydrophobicity, steric, and EIIP — 105 features per residue.
Clusters spatially adjacent interacting residues (Cα < 10Å) into epitope patches — regions antibodies actually bind.
Generates a .pml script for direct PyMOL loading — red/blue coloring with probability labels on every interacting residue.
Download Python standalone for local use. Supports multiple chains, RSA filtering, and custom probability thresholds.
Every prediction generates a complete set of files for downstream analysis and visualization.
Per-residue: Residue, RSA, Probability, Interacting/Non-interacting label.
Counts, percentages, average probability, and top 10 highest-scoring residues.
Probability in B-factor column. PyMOL: spectrum b, blue_white_red
Auto-colors red/blue with probability labels on Cα atoms of interacting residues.
Spatially clustered interacting residues with average patch probability scores.
Upload a PDB to get all 5 output files
HAIRpred2: Mehta N., et al. (2025) HAIRpred2: Structure-based prediction of antibody-interacting residues in human antigens. (manuscript in preparation)
HAIRpred (previous): Sahni R., Kumar N. and Raghava GPS (2025) HAIRpred: Prediction of human antibody interacting residues in an antigen from its primary structure. Protein Sci, 34(8):e70212