Code repository for "Improving Drug Target Interaction Prediction using Protein Language Models and Contact Maps".
PGraphDTA is a computational tool designed for predicting Drug-Target Interactions (DTIs) using advanced graph neural networks. This code is based on the research paper titled "Improving Drug Target Interaction Prediction using Protein Language Models and Contact Maps".
data_processing
: Contains scripts and utilities for pre-processing and preparing the data for training and inference.dti_inference_dist.py
: Script for DTI inference.dti_inference_dist_contact_map.py
: Script for DTI inference with contact map integration.dti_train_dist.py
: Script for training the DTI model.dti_train_dist_contact_map.py
: Script for training the DTI model with contact map integration.models
: Directory containing pre-trained models and architecture definitions.
- Clone this repository to your local machine.
- Set up a virtual environment (Anaconda3 recommended).
- Create environment: conda env create --file environment.yml.
To train the model, run:
python dti_train_dist.py [arguments]
For training with contact map integration, run:
python dti_train_dist_contact_map.py [arguments]
To infer using the trained model, run:
python dti_inference_dist.py [arguments]
For inference with contact map integration, run:
python dti_inference_dist_contact_map.py [arguments]
Ensure your data is placed in the appropriate directories and follows the expected formats. Refer to the data_processing
directory for utilities and scripts that can help in this regard.
The models
directory contains pre-trained models and their architecture definitions. You can use these for direct inference or as a starting point for further training.
If you find our repository helpful or used it, please cite our paper.
@misc{bal2023pgraphdta,
title={PGraphDTA: Improving Drug Target Interaction Prediction using Protein Language Models and Contact Maps},
author={Rakesh Bal and Yijia Xiao and Wei Wang},
year={2023},
eprint={2310.04017},
archivePrefix={arXiv},
primaryClass={cs.LG}
}