Bayesian models for screening and TB Mobile for target inference with Mycobacterium tuberculosis

Area of Science:

• Drug discovery and development
• Computational chemistry
• Infectious diseases

Background:

• High-throughput screening (HTS) in whole cells is standard for identifying anti-tubercular compounds.
• Traditional HTS methods often yield low hit rates, necessitating improved prioritization strategies.
• Mycobacterium tuberculosis remains a significant global health threat, driving the need for novel therapeutics.

Purpose of the Study:

• To employ Bayesian machine learning models for predicting anti-tubercular activity.
• To filter a large compound library to identify promising candidates for in vitro testing.
• To enhance the efficiency and hit rate of drug discovery for tuberculosis.

Main Methods:

• Utilized Bayesian machine learning models to predict anti-tubercular activity.
• Filtered an internal library of over 150,000 compounds.
• Selected and tested 48 compounds in vitro, achieving a 22.9% hit rate.
• Predicted potential targets for active compounds using TB Mobile and clustering.

Main Results:

• Identified 11 compounds with anti-tubercular activity (MIC values 0.4–10.2 μM).
• Achieved a high hit rate of 22.9%, significantly exceeding typical HTS rates.
• Discovered active compounds within quinolone series, long aliphatic linkers, and singleton groups.
• Predicted potential targets for the 11 active molecules.

Conclusions:

• Bayesian machine learning models effectively prioritize compounds for anti-tubercular drug discovery.
• This computational approach significantly increases hit rates compared to traditional HTS.
• The identified compounds and predicted targets offer avenues for further optimization and development of new tuberculosis treatments.