QSAR classification model for antibacterial compounds and its use in virtual screening.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Emerging drug-resistant bacterial strains necessitate the development of new antibacterial agents.
• Improving existing scaffolds and identifying novel ones are crucial for antibacterial drug discovery.

Purpose of the Study:

• To apply a Bayesian classification quantitative structure-activity relationship (QSAR) approach for rapid screening of chemical libraries.
• To identify compounds with predicted antibacterial activity.

Main Methods:

• Assembled datasets of 317 known antibacterial and diverse non-antibacterial compounds from PubChem Bioassays.
• Constructed a Bayesian classification model using structural fingerprints and physicochemical properties.
• Validated the model on an independent test set and screened ~200k compounds.

Main Results:

• Achieved 84% accuracy and 86% precision in identifying antibacterial compounds on an independent test set.
• Screened ~200k compounds, achieving up to ~76% accuracy and 1.5-2 fold enrichment of PubChem Bioassay actives.
• Identified both known and novel antibacterial scaffolds among the top screened hits.

Conclusions:

• A validated Bayesian classification QSAR approach can effectively complement other screening methods.
• This approach aids in the rapid identification of novel and promising antibacterial drug candidates.
• Publicly available datasets support model construction and validation for future research.