SpatConv is a powerful and highly efficient geometric deep learning model designed for identifying protein-protein interaction sites by leveraging the pre-trained protein language model ProtT5 and protein structure information. Users can predict protein-protein binding sites by submitting query proteins. Specifically, users need to input a PDB file containing the protein structure, and an h5 file extracted via a pretrained language model. After submission, the results page displays the predicted binding sites and scores for each residue of the query protein.

Readme: Please read the readme file carefully before using SpatConv.

Source code: https://github.com/gmnnnhh/SpatConv.

Datasets: https://github.com/gmnnnhh/SpatConv.

The example files are available below:
         SpatConv_example.zip

This work is openly licensed via CC0 1.0