Predicting Drug Resistance in Mycobacterium tuberculosis: A Machine Learning Approach to Genomic Mutation Analysis.

Area of Science:

• Genomics and Bioinformatics
• Infectious Disease Research
• Machine Learning in Medicine

Background:

• Tuberculosis (TB), caused by Mycobacterium tuberculosis, is a leading global infectious disease.
• Drug resistance in TB is a major challenge to effective treatment and control.
• Genomic sequencing and machine learning offer promising tools for predicting drug resistance.

Purpose of the Study:

• To evaluate four machine learning models for classifying resistance to ethambutol, isoniazid, and rifampicin in M. tuberculosis isolates.
• To compare model performance using different data preprocessing techniques, including PCA and mutation prioritization.

Main Methods:

• Trained Extreme Gradient Boosting Classifier (XGBC), Logistic Gradient Boosting Classifier (LGBC), Gradient Boosting Classifier (GBC), and Artificial Neural Network (ANN) models.
• Utilized a Variant Call Format (VCF) dataset preprocessed by the CRyPTIC consortium.
• Compared model performance on original, PCA-reduced, and mutation-prioritized datasets using sensitivity, specificity, Precision, F1-scores, and Accuracy.

Main Results:

• The XGBC model trained on the original dataset achieved the highest performance.
• Achieved high sensitivity (0.97, 0.90, 0.94), specificity (0.97, 0.99, 0.96), and F1-scores (0.93, 0.94, 0.92) for ethambutol, isoniazid, and rifampicin, respectively.
• Demonstrated superior accuracy of the XGBC model in classifying drug resistance compared to other models.

Conclusions:

• Binary representation of mutations effectively trains the XGBC model for predicting TB drug resistance and susceptibility.
• The XGBC model trained on the original dataset shows significant potential for clinical application in diagnosing drug-resistant TB.
• Further validation is recommended for broader implementation in TB diagnostics.