This repo contains the source code to run PeSTo and PeSTo-Carbs on PDB files.
To install dependencies, run
pip install -r requirements.txt
apply_model.py file can be used to run inference for binding site prediction.
| Argument | Required | Choices | Description |
|---|---|---|---|
--config_model_name |
Yes | ps-s, ps-g, pesto |
Model configuration name to use. |
--device |
No (default: cpu) |
cpu, cuda |
Device to run the model on. |
--output_folder |
Yes | - | Path to save model predictions. |
--input_folder |
Yes | - | Path containing .pdb files to process. |
model pesto can be used for prediction of protein, dna/rna, small molecule, ion and lipid binding interfaces. While using pesto the script will output five PDB files <pdbid>_i[0-4].pdb these are for predictions on protein-protein, protein-nucleic acid, protein-ion, protein-ligand, and protein-lipid, respectively. For carbohydrate binding site prediction I used the ps-g model. This will generate two PDB outputs <pdbid>_i0.pdb for protein-carbohydrate prediction and <pdbid>_i1.pdb for protein-cyclodextrin prediction. I only used the <pdbid>_i0.pdb
Example usage
python apply_model.py --config_model_name=pesto --device=cuda --output_folder=./output_pesto --input_folder=./inputdir
The predicted values are stored in the b-factor column. This can be visualized in PyMOL using:
spectrum b, blue_white_red, all, 0, 1
Or in ChimeraX using:
color bfactor palette "#2B59C3:#D1D1D1:#D7263D" range 0,1
process_interface_pred.py can be used to store per-residue predictions to a csv file. Please change the file paths accordingly.