This repo contains the code for our paper " GanDTI: a Multi-task Neural Network for Drug-Target Interaction Prediction"
by Shuyu Wang*, Peng Shan
we report GanDTI, an end-to-end deep learning model for both interaction classification and binding affinity prediction tasks. This model employs the compound graph and protein sequence data. It only consists of a graph neural network, an attention module and a multiple-layer perceptron, yet outperforms the state-of-the art methods on the DUD-E, human, and bindingDB benchmark datasets. This demonstrates our refined model is highly effective and efficient for DTI prediction and provides a new strategy for performance improvement.
- Python 3.7
- Pytorch
- numpy
- pickle
- RDKit
- sklearn
- CUDA
- to test human dataset: first
cd human
and then run
python dataProcess.py
to generate the files required. Then
cd ..
python main.py --dataset human --mode classification
- to test BindingDB Ki dataset: first
cd ki
and then run
python dataProcess.py
to generate the files required. Then
cd ..
python main.py --dataset ki --mode regression
- to use DUD-E dataset: first
cd dude
and then unzip the dataset file to process
python dataProcess.py
to generate the files required. Then
cd ..
python main.py --dataset dude --mode classification
We'd like to express our gratitude towards all the colleagues and reviewers for helping us improve the paper. The project is impossible to finish without the open-source implementation.