rdk/p2rank

What it solves

P2Rank is designed to predict ligand-binding sites (pockets) on protein structures. It provides a fast and accurate way to identify where a small molecule (ligand) is likely to bind to a protein without needing external software for complex feature calculation or a database of known protein-ligand templates.

How it works

The tool uses a machine learning model trained on known protein-ligand complexes. It scores and clusters points on the protein's solvent accessible surface (SAS) to determine the "ligandability" of individual points. It can also rescore pockets predicted by other tools (like Fpocket or DeepSite) to improve accuracy.

Who it’s for

It is primarily for researchers in computational chemistry, drug discovery, and structural biology who need to identify potential binding sites on proteins, including those derived from X-ray crystallography, AlphaFold models, NMR, or cryo-EM structures.

Highlights

• High Accuracy: Achieves higher identification success rates for top-ranked pockets compared to some traditional methods.
• Broad Format Support: Supports PDB, mmCIF, and BinaryCIF files, including gzipped and Zstandard compressed formats.
• Specialized Profiles: Includes a specific configuration for AlphaFold models and other modeled structures that ignores B-factor features.
• Integrated Rescoring: Can rescore predictions from a variety of other pocket-prediction tools.
• Visualizations: Generates scripts for PyMOL and ChimeraX to visualize predicted pockets.
• Export Options: Offers optional tabular exports for pocket descriptors (geometric, shape, electrostatic) and 3D grids of points.