DoGSiteScorer

DoGSiteScorer predicts and scores protein binding pockets to identify and assess binding site druggability using structural descriptors.

Key Features:

• Difference of Gaussian (DoG) detection: Uses a Difference of Gaussian approach derived from image processing techniques to identify potential binding pockets in protein structures.
• Active site prediction accuracy: Predicts binding pockets with over 92% success on benchmark datasets such as PDBBind and scPDB.
• Subpocket identification: Subdivides predicted pockets into smaller subpockets, with 63% of pockets in the PDBBind dataset found to be further divisible.
• Ligand coverage analysis: Reports that the cocrystallized ligand is contained within exactly one subpocket in 87% of cases, pockets encompass at least half of the ligand in 90% of cases, and more than a quarter of the pocket in 70% of cases.
• Druggability scoring: Computes an automated druggability score per (sub)pocket based on a linear combination of descriptors such as volume, hydrophobicity, and enclosure.

Scientific Applications:

• Druggability assessment: Identification and ranking of druggable targets in structure-based drug discovery projects through pocket detection and scoring.
• Structural analysis for target selection: Structural characterization of active sites and subpocket topology to inform protein function prediction, classification, and target selection.

Methodology:

Applies a Difference of Gaussian (DoG) approach to detect pockets, subdivides pockets into subpockets, computes descriptors (volume, hydrophobicity, enclosure) combined linearly for druggability scoring, and evaluates performance via benchmarks on PDBBind and scPDB including ligand coverage analyses.