PrankWeb

PrankWeb predicts ligand binding sites on protein structures to identify potential druggable pockets and support structural bioinformatics analyses.

Key Features:

• Template-Free Prediction: Uses the P2Rank algorithm to predict binding sites without relying on structural templates.
• Machine Learning-Based Approach: Employs machine learning to assign ligandability scores to points on protein surfaces.
• Sequence Conservation Analysis: Integrates sequence conservation data to refine and prioritize predicted binding sites.
• Export Options: Exports predicted binding sites as PyMOL scripts for offline visualization and analysis.
• REST API Integration: Provides a REST API for programmatic submission and retrieval to support high-throughput analyses.
• Speed and Efficiency: Multi-threaded implementation yields runtimes under one second per protein.

Scientific Applications:

• Protein function elucidation: Maps potential ligand-binding pockets to inform functional annotation of proteins.
• Structure-based drug discovery: Identifies and ranks druggable pockets to prioritize targets and guide ligand design.
• High-throughput structural bioinformatics: Enables large-scale, automated analyses through fast runtimes and API access.
• Benchmarking and comparative evaluation: Provides predictions that have been reported to outperform tools such as Fpocket, SiteHound, MetaPocket 2.0, and DeepSite.

Methodology:

Places points on the solvent-accessible surface of proteins, predicts ligandability of points using P2Rank's machine learning model based on local chemical neighborhoods, clusters high-scoring points into binding sites, and refines predictions with sequence conservation data; implemented as a multi-threaded algorithm with sub-second per-protein runtimes.