PocketDepth: a new depth based algorithm for identification of ligand binding sites in proteins

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Predicting functional sites in proteins is crucial for understanding protein function and for structure-based drug design.
• Existing methods often lack systematic approaches to identify these sites.
• Depth is an important, yet underutilized, parameter in protein structure analysis.

Purpose of the Study:

• To introduce PocketDepth, a novel geometry-based algorithm for predicting protein binding sites.
• To evaluate the performance of PocketDepth using different parameter sets ('deeper' and 'surface') against a benchmark dataset.
• To assess the potential of PocketDepth for advancing structural biology and drug discovery.

Main Methods:

• PocketDepth employs a geometry-based approach utilizing depth-based clustering to identify putative protein pockets.
• The algorithm was tested on the PDBbind dataset, comprising 1091 proteins.
• True-positive predictions were defined by at least 10% overlap with known ligand-binding sites.

Main Results:

• The 'deeper' parameter set achieved 77% prediction accuracy, with 55.2% in the first rank.
• The 'surface' parameter set yielded a high prediction rate of 95.8%, though with lower ranks.
• Combining both parameter sets resulted in 96.5% accuracy, with 41.8% in the first rank, and improved coverage in top ranks.

Conclusions:

• PocketDepth offers a robust and accurate method for predicting protein binding sites.
• The 'deeper' set excels at precise pocket boundary identification, while the 'surface' set provides broader coverage.
• The combined approach significantly enhances prediction accuracy and ranking, offering a valuable tool for structural biology and drug design.