DPubChem

DPubChem applies machine-learning QSAR modeling and high-throughput virtual screening to PubChem BioAssay and HTS datasets to predict active compounds and construct compound–assay interaction networks for drug discovery.

Key Features:

• Machine learning-based QSAR: Uses advanced machine-learning algorithms to derive quantitative structure–activity relationship models from assay data.
• HTS heterogeneous data analysis: Analyzes large-scale heterogeneous data from high-throughput screening (HTS) to identify activity signals across assays.
• PubChem BioAssay integration: Processes and analyzes datasets from the PubChem BioAssay database to inform model training and prediction.
• Performance metrics: Demonstrated prediction across 300 datasets with an average geometric mean of 76.68% and an average F1 score of 76.53%.
• Interaction network construction: Constructs compound–assay interaction networks to reveal predicted links between chemical compounds and biological assays.

Scientific Applications:

• Novel drug candidate identification: Prioritizes and suggests candidate compounds for follow-up based on QSAR predictions and interaction networks.
• Disease-specific candidate discovery: Supported identification of a potential treatment candidate for Niemann-Pick type C disease using integrated QSAR and screening data.

Methodology:

Application of advanced machine-learning algorithms to derive QSAR models from extensive datasets in the PubChem BioAssay database, analysis of large-scale heterogeneous HTS data, and construction of compound–assay interaction networks.