Machine learning-based prediction of antibiotic resistance in Mycobacterium tuberculosis clinical isolates from Uganda

The study explores the use of machine learning algorithms to predict drug resistance of four anti-TB drugs in Mycobacterium tuberculosis (MTB) using whole-genome sequence and clinical data from Uganda and South Africa. The best model was selected based on the highest Mathews Correlation Coefficient (MCC) and Area Under the Receiver Operating Characteristic Curve (AUC) score. Logistic regression excelled in predicting rifampicin and streptomycin resistance, while Extreme Gradient Boosting (XGBoost) and Gradient Boosting (GBC) for ethambutol and Gradient Boosting (GBC) for isoniazid. HIV status was identified as a significant feature in predicting drug resistance. The study suggests that integrating diverse data types, such as genomic and clinical data, could improve resistance predictions, support robust surveillance systems, and inform targeted interventions to curb the rising threat of antimicrobial resistance.