DPubChem: a web tool for QSAR modeling and high-throughput virtual screening.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• High-throughput screening (HTS) identifies active compounds but faces challenges with large, imbalanced datasets.
• Quantitative structure-activity relationship (QSAR) models offer faster virtual screening but often have limitations like high false positive rates.

Purpose of the Study:

• To develop DPubChem, a novel web tool for deriving accurate QSAR models using advanced machine learning techniques.
• To enable efficient analysis of experimental data from the PubChem BioAssay database.
• To create interaction networks for predicting novel links between compounds and biological assays.

Main Methods:

• Implemented state-of-the-art machine learning techniques within the DPubChem web tool.
• Utilized the PubChem BioAssay database for deriving and validating QSAR models.
• Developed interaction network analysis to predict compound-assay relationships.

Main Results:

• DPubChem achieved an average geometric mean of 76.68% and F1 score of 76.53% across 300 datasets.
• The tool successfully predicted active compounds, enhancing model precision.
• DPubChem identified a novel drug candidate for Niemann-Pick type C disease through network analysis.

Conclusions:

• DPubChem provides a user-friendly platform for accurate QSAR modeling and drug discovery.
• The tool's interaction networks facilitate the discovery of novel therapeutic applications.
• DPubChem represents a significant advancement in leveraging large-scale bioactivity data for drug development.