GeoNet

GeoNet is an interpretable geometric deep learning model for predicting DNA, RNA, and protein binding sites by learning the latent residue binding patterns.

Readme: We provide two modes of GeoNet, Fast and the Standard, please read the readme file carefully before using them. Users can choose the fast and the standard mode of the GeoNet, the former just need to upload the protein structure file, while the latter need users to upload the pssm and hmm file for each protein chain. And for the protein-DNA and -RNA data, we only use the monomer as the training data.

Note: The uploaded pdb file must be named as {PDBID}_{chainID}.pdb!!!

Source code: https://github.com/hjy23/GeoNet.

Datasets: https://zenodo.org/records/13743762.

The input files of the case study in the paper are available below:
GeoNet_case.zip

This work is openly licensed via CC0 1.0

Upload a PDB file (Note: the name of the file comprises the pdbID and chain(s), e.g., 6ide_B.pdb)

Choose the ligand type

DNA

RNA

Protein

Choose the model mode

Fast mode

Standard mode

Upload the pssm File corresponding to each chain (e.g., 6ide_B.pssm). You can run PSI-BLAST against the Uniref90 protein database to generate PSSM file.

Upload the hhm File corresponding to each chain (e.g., 6ide_B.hhm). You can run HHblits against the Uniclust30_2018_08 protein database to generate HMM file.