PocketDepth

PocketDepth predicts functional binding sites (pockets) within protein structures using a geometry-based, depth-driven approach to quantify pocket centrality and support structural analysis and structure-based drug design.

Key Features:

• Geometry-based algorithm: Employs a geometry-based framework to identify and delineate putative pockets in protein structures.
• Depth parameter quantification: Quantifies "depth" to measure how central a given subspace is within a pocket.
• Depth-based clustering: Uses depth-based clustering to systematically evaluate subspace centrality within putative pockets.
• Parameter sets ('deeper' and 'surface'): Provides two parameter sets—'deeper' for precise pocket-boundary identification and 'surface' for broader coverage of potential binding sites.
• 'Deeper' performance: 'Deeper' parameters achieve 77% accuracy for any ranked prediction, 55.2% first-ranked, 91.2% within top five, and 97.4% within top ten.
• 'Surface' performance: 'Surface' parameters prioritize coverage and yield a 95.8% overall prediction rate.
• Combined-parameter strategy: Algorithmically combines 'deeper' and 'surface' parameters, producing 96.5% accuracy for any ranked prediction, ~41.8% first-ranked, 82% within top five, and 94% within top ten.
• Benchmark dataset: Performance was evaluated against PDBbind, comprising 1,091 curated protein structures.

Scientific Applications:

• Protein functional-site prediction: Identification and ranking of likely ligand-binding pockets in protein structures.
• Structure-based drug design: Prioritization of candidate binding sites to inform ligand placement and lead discovery.
• Protein–ligand interaction analysis: Assessment of pocket centrality and geometry to support analysis of protein–ligand interfaces.
• Method benchmarking and validation: Validation and comparative evaluation of pocket-prediction approaches using the PDBbind dataset.

Methodology:

Geometry-based depth quantification and depth-based clustering were used to evaluate subspace centrality within putative pockets; two parameter sets ('deeper' and 'surface') were applied and algorithmically combined, and performance was validated on the PDBbind dataset of 1,091 curated structures.