🧬 ImaPEp: Antibody-Antigen Binding Probability Prediction

ImaPEp is a computational tool for predicting the binding probability between antibody paratopes and antigen epitopes. This repository provides the necessary scripts for training models and performing predictions on antibody-antigen complexes using structural data.

For full details, please refer to our publication: D.Li, F.Pucci, M.Rooman “Prediction of paratope–epitope pairs using convolutional neural networks”.

🛠️ Dependencies

ImaPEp has been tested on Linux with the following Python packages:

• PyTorch == 1.13.0
• TorchVision == 0.14.0
• Numpy == 1.25.2
• abnumber
• tqdm
• matplotlib == 3.7.2
• Biopython == 1.78
• Scikit-learn == 1.3.0
• Pillow == 10.0.1

▶️ Usage

To use ImaPEp, first clone the repository and update your environment variables:

cd /your/path/
git clone https://github.com/3BioCompBio/ImaPEp.git 
export PYTHONPATH="/your/path/ImaPEp/:$PYTHONPATH"

To make the changes permanent, add the export command to your ~/.bashrc file.

Training a Model from Scratch

To train ImaPEp from scratch using your dataset, run the following command:

python3 train.py --train-dir=train_resi --model-dir=models

This will train ten cross-validated models using the dataset in train_resi and save the trained model parameters in the models directory.

Predicting Antibody-Antigen Binding

ImaPEp supports two modes for prediction:

1. Single-File Mode

For predicting the binding affinity of a single antibody-antigen complex, provide a PDB structure as input:

python3 predict.py --sample-id=my_pdb --input=my_pdb.pdb --chains=HLBC

Here, the --chains argument specifies the relevant protein chains in the PDB file:

• The first character represents the antibody heavy chain ID (e.g., H).

• The second character represents the antibody light chain ID (e.g., L).

• The remaining characters correspond to the antigen chains (e.g., BC).

2. Batch mode

For processing multiple antibody-antigen complexes in one go, place all input PDB files into a designated folder and create a job.txt file containing the details for each complex (one per line in the format pdb_id,chains). Then, execute:

python3 predict.py --input=/job/folder/

The chains column in the job.txt file follows the same format as in Single-File Mode, specifying the heavy chain, light chain, and antigen chains.

📝 Notes

• The antigen should be a protein with at least 50 residues.

• The paratope and epitope should each contain at least three residues.

• All input PDB files must have no missing atoms. If residues with missing atoms exist, consider removing them or adding the missing atoms (recommended).

• In rare cases, the CDR regions may not be successfully parsed, causing the prediction to terminate.

🔗 Citation

Please cite our publication when using ImaPEp in your research:

@article{li2024imapep,
    title    = {Prediction of paratope–epitope pairs using convolutional neural networks},
   author    = {Li, D. and Pucci, F. and Rooman, M.},
   journal   = {International Journal of Molecular Sciences},
   year      = {2024}
}