Prediction of Pyrazinamide Resistance in <i>Mycobacterium Tuberculosis</i> Using Structure-Based Machine Learning Approaches

Abstract

Pyrazinamide is one of four first-line antibiotics used to treat tuberculosis. While phenotypic antibiotic susceptibility testing for pyrazinamide is problematic, genetic variation in pncA drives pyrazinamide resistance in clinical isolates. Using a derivation dataset of 291 non-redundant, missense pncA mutations with high-confidence phenotypes, we trained machine learning models to predict pyrazinamide resistance based on sequence- and structure-based features. The models were further benchmarked by predicting the pyrazinamide resistance phenotype of 2,292 clinical isolates harboring pncA missense mutations. The probabilities of resistance predicted by the model were compared with in vitro pyrazinamide minimum inhibitory concentrations of 71 isolates to determine whether the machine learning model could predict the degree of resistance. This capacity of this approach to predict the effects of all pncA missense mutations improves the sensitivity and specificity of pyrazinamide resistance prediction in genetics-based clinical microbiology workflows for tuberculosis and provides a proof-of-concept for other drugs.