Prediction of QcrB Inhibition as a Measure of Antitubercular Activity with Machine Learning Protocols

Area of Science:

• Medicinal Chemistry
• Computational Biology
• Drug Discovery

Background:

• Tuberculosis (TB) treatment faces challenges due to Mycobacterium tuberculosis's rapid mutation and dormancy.
• The cytochrome bcc complex (QcrB) presents a novel therapeutic target for TB.
• Q203 is the first molecule identified targeting QcrB.

Purpose of the Study:

• To employ machine learning (ML) approaches for designing novel molecules targeting the QcrB protein.
• To develop and optimize ML models for predicting QcrB inhibitors.
• To validate the predictive accuracy of ML models using known compounds.

Main Methods:

• Development of ML models using a dataset of 350 molecules.
• Utilized three molecular feature sets: MACCS keys, ECFP6 fingerprints, and Mordred descriptors.
• Trained and optimized eight ML classifier algorithms for each feature set.

Main Results:

• The support vector machine classifier with ECFP6 fingerprints demonstrated the highest accuracy.
• The developed ML model successfully classified known potent imidazopyridine amide inhibitors as active.
• Model validation confirmed its capability in identifying effective QcrB inhibitors.

Conclusions:

• ML-based approaches are effective for designing novel anti-tubercular agents targeting QcrB.
• The optimized SVM model with ECFP6 features shows potential for future drug discovery efforts.
• This study provides a validated computational framework for identifying new QcrB inhibitors.