GALA

Overview

We introduce GALA, a versatile graph learning framework designed to decipher the intricate mechanisms of antigen-antibody interactions. By leveraging a structure-aware protein language model and invariant geometric representations, GALA provides a hierarchical approach to decoding molecular recognition, bridging the gap between identifying general protein binding sites, mapping antigen epitopes, and quantifying binding affinities.

Install requirements

SaProt

Please follow the instructions in SaProt to install and use the SaProt_650M_PDB model to generate embeddings. The installed SaProt model and DSSP tool should be put in ./model.

ANARCI

Please download the ANARCI for numbering amino-acid sequences of antibody and identifying the CDR sites.

Environments

Required dependencies and versions:

• python 3.8.16
• pytorch 1.13.1
• biopython 1.79
• transformers 4.39.3
• e3nn 0.5.1
• numpy 1.23.5
• torch-geometric 2.3.0
• pytorch-scatter 2.1.1
• pytorch-cluster 1.6.1

Datasets

• The PPBS dataset: the protein-protein binding site dataset used for pretraining was obtained from PeSTo.
• The Epi_GB and Epi_SN datasets: the antibody-agnostic epitope datasets Epi_GB and Epi_SN were obtained from GraphBepi and ScanNet, respectively.
• The Epi_SAb and Aff_SAb dataset: the antibody-specific epitope dataset (Epi_SAb) and antibody-antigen binding affinity dataset (Aff_SAb) used in our study are available in ./Antibody_specific_epitope/data and ./Antibody_antigen_affinity/data, respectively. All complex structures can be downloaded from SAbDab.

Prediction

Antibody-agnostic Epitope Prediction

Inference commands using trained models (located in ./Antibody_agnostic_epitope/model):

python test.py --task antigen --data-path ./Antibody_agnostic_epitope/examples --antigen-name 1a2y_C --gpu 0 --out ./output/antigen

Antibody-specific Epitope Prediction

Inference commands using trained models (located in ./Antibody_specific_epitope/model):

python test.py --task pair_epitope --data-path ./Antibody_specific_epitope/examples --csv ./Antibody_specific_epitope/examples/example.csv --pdb ./Antibody_specific_epitope/examples/pdb --gpu 0 --out ./output/specific_epitope

Antibody-antigen Affinity Prediction

Inference commands using trained models (located in ./Antibody_antigen_affinity/model):

python test.py --task pair_affinity --data-path ./Antibody_antigen_affinity/examples --csv ./Antibody_antigen_affinity/examples/example.csv --pdb ./Antibody_antigen_affinity/examples/pdb --gpu 0 --out ./output/affinity