Machine Learning Models for Mycobacterium tuberculosisIn Vitro Activity: Prediction and Target Visualization.

Area of Science:

• * Medicinal Chemistry and Computational Biology
• * Drug Discovery and Development

Background:

• * Tuberculosis (TB), caused by Mycobacterium tuberculosis (Mtb), remains a critical global health issue with high mortality rates.
• * Despite extensive screening, limited investment hinders the development of new TB drugs, necessitating innovative approaches.
• * Existing drug discovery pipelines face challenges in efficiently identifying novel Mtb-targeting compounds.

Purpose of the Study:

• * To develop and validate machine learning (ML) models for accelerated discovery of novel anti-TB compounds.
• * To identify key molecular features associated with Mtb activity.
• * To create publicly accessible datasets and models for Mtb drug discovery.

Main Methods:

• * Curated a large dataset of small-molecule Mtb data (18,886 molecules).
• * Developed and validated Bayesian classification models using this curated data.
• * Utilized ML models for compound library scoring, molecular design, and target identification.

Main Results:

• * Validated ML models using an independent library of over 1000 synthesized molecules.
• * Identified specific molecular features enriched in active anti-Mtb compounds.
• * Generated new regression and classification models for compound evaluation and design.

Conclusions:

• * Machine learning significantly enhances the efficiency of Mtb drug discovery.
• * Publicly released datasets and models will foster further research and development of new TB therapeutics.
• * The study provides valuable tools and insights for identifying novel anti-TB drug candidates.