/GrASP

Graph Attention Site Prediction (GrASP): Identifying Druggable Binding Sites Using Graph Neural Networks with Attention

Primary LanguagePythonMIT LicenseMIT

Graph Attention Site Prediction (GrASP)

Preprint

https://biorxiv.org/content/10.1101/2023.07.25.550565v1

Colab

Fetch a PDB file and try GrASP on it in our Colab demo.

Open in Collab

Download Datasets

Coming soon!

How to Run

Currently, only production mode on a pre-trained model is supported until datasets are online.

  • Build the conda environments in ./envs/ob_env.yml and ./envs/pytorch_env.yml. This will add two new conda environments named ob and pytorch_env respectively.
conda env create -f envs/ob_env.yml
conda env create -f envs/pytorch_env.yml
  • Move protein structures to ./benchmark_data_dir/production/unprocessed_inputs/. Heteroatoms do not need to be removed, they will be cleaned during parsing.
  • Load ob and parse the structures into graphs.
conda activate ob
python3 parse_files.py production
  • Run GrASP over the protein graphs.
conda deactivate; conda activate pytorch_env
python3 infer_test_set.py
  • Paint structures with GrASP predictions in the b-factor column.
conda deactivate; conda activate ob
python3 color_pdb.py

Supported Formats

PDB and mol2 formats are supported and validated. Other formats supported by both MDAnalysis and OpenBabel 2.4.1 may be working but have not been tested.