Toward a benchmarking data set able to evaluate ligand- and structure-based virtual screening using public HTS data.

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Bioinformatics and cheminformatics

Background:

• Virtual screening accelerates drug discovery but requires robust validation datasets.
• Existing datasets often suffer from analogue bias and artificial enrichment.
• Public databases like PubChem BioAssay offer vast resources for creating new validation sets.

Purpose of the Study:

• To identify and curate suitable PubChem BioAssay datasets for validating virtual screening methods.
• To create novel validation datasets with realistic ratios of active and inactive compounds.
• To benchmark the utility of these datasets using structure- and ligand-based screening approaches.

Main Methods:

• Screening PubChem BioAssay database for high-throughput screening data.
• Filtering for assays with available target crystal structures in the Protein Data Bank (PDB).
• Compiling seven curated datasets (MMP13, DUSP3, PTPN22, EPHX2, CTDSP1, MAPK10, CDK5) for validation.

Main Results:

• Seven novel datasets were compiled, featuring a high number of inactive compounds relative to active ones (19-369 active, 59,405-337,634 inactive).
• Initial performance evaluation using docking and 3D shape similarity was conducted.
• Dataset characterization involved physicochemical similarity and 2D fingerprint analyses.

Conclusions:

• The newly compiled datasets provide a valuable resource for evaluating virtual screening methods.
• These datasets address limitations of existing benchmark sets by offering more realistic compound ratios.
• They can serve as a useful complement for assessing the performance of computational drug discovery tools.