A CatBoost machine learning for prognosis of pathogen’s drug resistance in pulmonary tuberculosis

Drug resistance of Mycobacterium tuberculosis is one of the modern challenges for the global public health system. Timely detection of resistant strains is crucial for decision of surgery or therapy prolongation. The routine procedure requires extensive experiments with isolated bacterial culture in specially equipped labs only. In contrast we propose a diagnostic tool based on studying special biochemical markers from a blood test that is realizable in usual clinics. To make a decision, we explored the predicting capacity of the machine learning with a modern effective algorithm CatBoost trained with the clinical data. These data have been obtained from patients with pulmonary tuberculosis forced by different strains of mycobacteria. The revealed most influential training biomarkers and parameters assure accuracy acceptable for a primary diagnostics.