Machine learning approaches to predict drug resistance in tuberculosis

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Tuberculosis (TB) poses a global health challenge, exacerbated by drug resistance.
• Traditional TB diagnostics are slow, costly, and lack accuracy.
• Genomic variants offer a potential avenue for improved TB drug resistance prediction.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting drug resistance in Mycobacterium tuberculosis.
• To investigate the efficacy of ensemble ML models using sequence and structure-based genomic features.
• To establish a gene-specific strategy for optimizing resistance prediction.

Main Methods:

• Compiled a dataset of TB mutations and resistance phenotypes from multiple databases.
• Extracted sequence-based (e.g., physicochemical properties, Provean scores) and structure-based features for each mutation.
• Evaluated ensemble ML models (Stacking, Bagging, Voting Classifiers) for predicting resistance to key anti-TB drugs.

Main Results:

• Model performance varied across six TB resistance genes (gyrA, gyrB, inhA, katG, rpoB, pncA).
• Accuracy ranged from 66% (gyrA Stacking) to 91.37% (pncA Voting); ROC scores from 0.69 to 0.92.
• Optimal models were gene-specific: Bagging for gyrA, gyrB, rpoB; Stacking for inhA; Voting for katG, pncA.

Conclusions:

• Gene-specific ensemble ML models utilizing comprehensive genomic features can effectively predict M. tuberculosis drug resistance.
• This approach offers a faster and more accurate diagnostic potential.
• Findings are a proof-of-concept requiring validation on larger clinical datasets.