Sequence-Only Prediction of Binding Affinity Changes: A Robust and Interpretable Model for Antibody Engineering

Introduction

ProtAttBA is a protein language model that predicts binding affinity changes based solely on the sequence information of antibody-antigen complexes.

Usage

Install

1. Create conda environment

conda create -n protab python==3.10

2. Install environment dependency

# activate environment
source activate protab
# install pytorch
conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=11.8 -c pytorch -c nvidia 
(or use pip: pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url https://download.pytorch.org/whl/cu118)

# install dependencies
pip install -r ./requirments.txt

dataset

Cross validation dataset is located in the cross_validation/data/csv folder (Source: Jin et al., 2024 ）

Sequence identity dataset is located in the seq-identity_sig-mul/data/identity_data folder (Use MMseqs with --min-seq-id 0.3)

Single mutation training and multi-mutation testing dataset is located in the seq-identity_sig-mul/data/sigmul_data folder

Training

# For cross validation you can use cross_validation/scripts/bash_cross-validation.sh with different args
cp bash_cross-validation.sh ../
bash bash_cross-validation.sh 

# For Sequence identity you can use seq-identity_sig-mul/scripts/bash_seq_identity.sh with different args
cp bash_seq_identity.sh ../ 
bash bash_seq_identity.sh

# For Single mutation training and multi-mutation testing you can use seq-identity_sig-mul/scripts/bash_seq_sigmul.sh with different args
cp bash_seq_sigmul.sh ../ 
bash bash_seq_sigmul.sh

Evaluation

# For evaluation you can use the seq-identity_sig-mul/eval.py to predict the result by change the args
python eval.py