Peptide Pose Predictor

This project provides a tool to predict the quality (RMSD) of a peptide pose docked to a protein using an EGNN model.

Installation

1. Clone the repository.

2. Install the package and dependencies:

   pip install .

   Note: For torch and torch_geometric, it is recommended to install them manually first to ensure compatibility with your CUDA version (if applicable).

Usage

The installation provides a command-line tool ppp.

ppp --prot <path_to_protein.pdb> --pep <path_to_peptide.pdb>

Arguments

• --prot: Path to the protein PDB file.
• --pep: Path to the peptide PDB file.
• --model: (Optional) Path to the trained model file (best_model_egnn.pth). If not provided, the script attempts to find it in the installation directory or current directory.

Example

ppp --prot examples/protein.pdb --pep examples/peptide.pdb

Files

• peptide_pose_predictor/predict.py: The main entry point for the CLI.
• peptide_pose_predictor/build_graph.py: Logic for converting PDB files to graphs.
• peptide_pose_predictor/inference.py: Inference logic and model definition.
• peptide_pose_predictor/training.py: Training script.
• peptide_pose_predictor/best_model_egnn.pth: Pre-trained model weights.