Struct2Query: Curated SiteHopper Database and Benchmark Datasets for Structure-Based Composite Query Generation

Dataset accompanying the Struct2Query method for structure-based composite query generation in virtual screening. Overview Struct2Query bridges structure-based and ligand-based virtual screening by leveraging protein binding site similarity to construct composite-molecule ROCS queries compatabile with rapid ligand-based screening. Given a target protein structure, it identifies proteins with similar binding pockets, retrieves their co-crystallized ligands, and assembles multi-molecule queries that capture binding modes across structurally related sites. This dataset provides the associated curated databases and benchmark data. Contents 1. SiteHopper Database (`sitehopper_databases/Struct2Query.shdb`) Curated database of 78,806 protein-ligand complexes Formatted for OpenEye SiteHopper pocket similarity searches 2. DEKOIS 2.0 Benchmark (`dekois/`) 81 protein targets with prepared structures Each target includes: protein structure (PDB), bound ligand (MOL2), active compounds (SDF), and decoys (SDF) 3. DUDE-Z Benchmark (`dudez/`) 43 protein targets Each target includes: active ligands (SMILES) and property-matched decoys (SMILES) Usage See the Struct2Query GitHub repository for installation instructions and usage examples.