This is the implementation of ZeroBind: A protein-specific zero-shot predictor with subgraph matching for drug-target interactions.
ZeroBind is built on Python3, we recommend using a virtual conda environment as enviroment management for the installation of ZeroBind and its dependencies. The virtual environment can be created as follows:
conda create -n your_environment python==3.9
conda activate your_environment
Download the source code of ZeroBind from GitHub:
git clone https://github.com/myprecioushh/ZeroBind.git
Install ZeroBind dependencies as following:
pip install pytorch==2.0.0 torchvision==0.15.0 torchaudio==2.0.0
conda install pyg==2.3.0 -c pyg
pip install lightning==2.0.1
pip install rdkit
pip install graphein
pip install fair-esm
pip install wandb
Multiple hyperparameters can be selected in meta.py.
python metaentry.py --batch_size=4 --num_workers=16 --num_inner_steps=5 --k_query=50
After model training starts, the progress bar will be automatically shown on your command line, and the trained model parameters will be saved in "checkpoints" dictory for every epoch.
python metaentry.py --test --num_workers=16 --k_query=50 --checkpoint_path="your_model_path"
After predicting with your well trained model, the predicting output will be saved in a "txt" file and the prediction metrics will be shown on your command line.
Online retrieval service and benchmark datasets are in (http://www.csbio.sjtu.edu.cn/bioinf/ZeroBind/).
This project is covered under the Apache 2.0 License.