MVGNN is a state-of-the-art graph neural network designed for predicting protein-protein binding sites with high speed and accuracy.
The MVGNN web server is available for free access here.
python 3.9.19
torch 1.13.1
cuda 11.7
transformers 4.41.1
Bio 1.83
Please visit ProtTrans on GitHub to download the model. Store the downloaded model in ./process_feature/pretrained_model/Rostlab/prot_t5_xl_uniref50.
1. Our protein structure predictions are generated using AlphaFold3. For detailed installation instructions, please refer to AlphaFold3 on GitHub.
2. We use the DSSP tool to extract secondary structure information. For detailed instructions, please refer to DSSP on GitHub. Before using the software, please grant execution permissions with:
chmod +x ./process_feature/Software/dssp-2.0.4/mkdssp
After completing the steps above, run the following command to generate features and save them in the ./feature directory:
python ./process_feature/process_feature.py
Alternatively, you can download the features directly from our cloud drive: Google Drive Link.
To predict datasets located in the ./datasets directory, run the following command:
python ./main.py
Ensure all dependencies are installed and the data is correctly formatted before executing the command.