Combining computational methods for hit to lead optimization in Mycobacterium tuberculosis drug discovery

Area of Science:

• Drug discovery and development
• Computational chemistry
• Machine learning in pharmacology

Background:

• Tuberculosis (TB) treatment requires shorter regimens and strategies to overcome drug resistance.
• Previous high-throughput screening identified numerous active and inactive compounds against Mycobacterium tuberculosis (Mtb).
• Computational modeling offers a promising approach to refine compound selection and minimize experimental testing.

Purpose of the Study:

• To develop and validate computational models for prioritizing antitubercular compounds.
• To assess the efficacy of clustering and Bayesian machine learning in enriching hit rates for drug screening.
• To identify non-toxic active compounds for further lead optimization.

Main Methods:

• Applied a cheminformatics clustering approach to analyze Mtb screening data.
• Utilized Bayesian machine learning models trained on public Mtb screening data.
• Selected and evaluated 1924 commercially available molecules for antitubercular activity and cytotoxicity across multiple cell lines (Vero, THP-1, HepG2).

Main Results:

• Achieved hit rates of 4.3% (Vero), 4.2% (THP-1), and 2.7% (HepG2) for antitubercular activity.
• Demonstrated significant enrichment of non-toxic active compounds using models incorporating cytotoxicity data.
• The combined model identified approximately 10% of hits in the top 1% screened, achieving over 10-fold enrichment.
• Successfully predicted known active compounds from academic and pharmaceutical screens.

Conclusions:

• The integration of clustering and Bayesian models is an effective strategy for prioritizing compounds in antitubercular drug discovery.
• This approach facilitates hit-to-lead optimization by identifying promising, non-toxic drug candidates.
• The validated models can accelerate the development of novel treatments for tuberculosis.