Randomized trial evaluates machine learning for predicting drug resistance mutations in drug-resistant tuberculosis, suggesting new treatment strategies.
Key Points
This research aims to develop a machine learning model to analyze mutations associated with drug resistance in Mycobacterium tuberculosis.
Developed a machine learning prediction model for drug-resistance mutations.
Analyzed 3,065 cases of drug-resistant tuberculosis using eight supervised algorithms.
Utilized SHapley Additive exPlanations for feature importance to identify significant mutations.
Random Forest classifier showed higher predictive performance than seven other algorithms with 10-fold cross-validation.
Identified mutations such as rpoB-I480T and gyrA-D94V, which may represent potential resistance markers.
Post-prediction analysis revealed novel mutations not found in WHO catalogues, warranting further investigation.