HAIRpred2

HAIRpred2 is a structure-based standalone tool for predicting antibody-interacting residues in antigen structures. It uses Relative Solvent Accessibility (RSA) combined with physicochemical properties in a sliding window framework, fed into a pre-trained Random Forest model.

pip install numpy pandas joblib gemmi biopython scipy

HAIRpred2 uses DSSP to compute RSA values. You need mkdssp installed on your system.

Linux / Ubuntu:

sudo apt install dssp

Conda (any platform):

conda install -c salilab dssp

macOS (Homebrew):

brew install dssp

git clone https://github.com/raghavagps/hairpred2
cd hairpred2
!wget https://webs.iiitd.edu.in/raghava/hairpred2/download/model
pip install numpy pandas joblib gemmi biopython scipy

Make sure best_model_random_forest.pkl is in the same folder as hairpred2.py.

Folder structure:

hairpred2/
├── hairpred2.py
├── best_model_random_forest.pkl
└── README.md

# Basic usage
#download the model first
!wget https://webs.iiitd.edu.in/raghava/hairpred2/download/model
python hairpred2.py -i antigen.pdb -c A

# Custom output prefix
python hairpred2.py -i antigen.pdb -c A -o my_results

# Multiple antigen chains
python hairpred2.py -i antigen.pdb -c A,B

# Filter out buried residues (RSA < 0.05)
python hairpred2.py -i antigen.pdb -c A --min-rsa 0.05

# Custom probability threshold
python hairpred2.py -i antigen.pdb -c A -t 0.4

All output files share the same prefix (set with -o):

Example:

Residue,RSA,Probability,Prediction
T22,0.1823,0.3812,Non-interacting
N23,0.6541,0.6247,Interacting
S24,0.7102,0.7103,Interacting
K25,0.0431,0.2891,Non-interacting

# After running the tool, open PyMOL and run:
@hairpred2_results.pml

This colors interacting residues red and non-interacting residues blue, with a surface overlay on the predicted epitope.

# PyMOL
load hairpred2_results_bfactor.pdb
spectrum b, blue_white_red

# ChimeraX
open hairpred2_results_bfactor.pdb
color bfactor

1. Validate — checks PDB file, ATOM records, and chain availability
2. Extract — isolates the antigen chain(s) using gemmi
3. RSA — computes per-residue Relative Solvent Accessibility via DSSP
4. Features — builds a 105-element feature vector per residue (7 features × 15 window positions)
5. Predict — runs the pre-trained Random Forest and outputs interaction probabilities
6. Epitope patches — clusters spatially proximal interacting residues (Cα distance < 10 Å)
7. Save — writes all 5 output files

HAIRpred2 is also available as a web server: https://webs.iiitd.edu.in/raghava/hairpred2/

If you use HAIRpred2 in your work, please cite our paper (citation to be added upon publication).

https://doi.org/10.5281/zenodo.19876445