This project requires ProteinGNN to parse pdb to PyG compactible format. Please follow the installation process there.
To build the datasets, place all AlphaFold2 structures under data/alphafold2/your_dataset, fasta under data/fasta
and experiment csv under data/csv.
python build_dataset.py --embedding esm --radii 6 --dataset your_dataset --n_processes N_PROCESSES
To train sequence-only and geometric models and visualize their performances,
bash batch_train.sh
python compare_supervised.py --rootdir esm-6
To further compare with unsupervised predictions, place pssm files under data/pssm and run
python ESM.py --stage preprocess --esm_install_path ESM_INSTALL_DIR
bash benchmark_esm.sh
python ESM.py --stage postprocess
python compare_unsupervised.py
For embedding locality analysis,
python locality.py