FINDSITEsupcomb2.0

FINDSITEsupcomb2.0 performs virtual ligand and target screening of proteins and biomolecules to predict protein–ligand interactions for drug lead discovery and target identification.

Key Features:

• Threading/structural-based approach: Employs a threading/structural-based methodology that enables virtual screening using low-resolution predicted structures and is faster and more accurate than high-resolution structure-based docking for large-scale screenings.
• Template library parsing: Parses known protein–ligand interactions from PDB, DrugBank, and ChEMBL for template ligand selection.
• Domain-aware template division: Divides template proteins into domains to prevent selection of ligands with falsely matched domains.
• Structure-comparison thresholds: Applies various thresholds during structure comparison to filter out inaccurately matched templates.
• Performance benchmarking: Benchmarking against the DUD-E set increased the 1% enrichment factor from 16.7 to 22.1 at a 30% sequence identity cutoff (p-value = 4.3 × 10^-3) and achieved a 1% ROC enrichment factor of 52.39 at an 80% sequence identity cutoff, outperforming deep convolutional neural network methods.
• Seed-ligand and high-resolution independence: Operates without requiring high-resolution structural data or a predefined set of seed ligands for virtual screening.

Scientific Applications:

• Drug lead discovery: Enables identification of potential drug leads through virtual ligand screening using low-resolution predicted structures.
• Drug target identification: Supports virtual target screening of biomolecules to determine likely drug targets.

Methodology:

Uses threading/structural-based alignment on low-resolution predicted structures, parses PDB, DrugBank, and ChEMBL for template ligands, divides template proteins into domains, applies thresholds during structure comparison, and was benchmarked on the DUD-E set with 30% and 80% sequence identity cutoffs.