Predicting tuberculosis drug resistance using machine learning based on DNA sequencing data

Abstract

Tuberculosis is a serious infectious disease caused by Mycobacterium tuberculosis (MTB) that primarily affects the lungs. It is known that several strains of MTB are resistant to drugs used in the treatment. This situation calls for the importance to detect and prevent further drug resistance and thus reducing the mortality rate. The conventional molecular diagnostic test is costly, requires a long time to conduct, and has low prediction ability. This research aims to explore the Machine Learning approach to accurately predict drug resistance which offers a much faster and cheaper solution than the conventional one. Experiments were carried out on 3393 isolates of MTB using several Machine Learning algorithms including C4.5, Random Forest, and Logitboost. Multiple drugs evaluated in this study include rifampicin (RIF), isoniazid (INH), pyrazinamide (PZA), and ethambutol (EMB). By using 10-fold cross-validation, the result had demonstrated that the model can accurately predict drug resistance with an accuracy of 99% and with Area Under Curve (AUC) reaching (near) 1. This result suggests that Machine Learning approach has a promising result in predicting Tuberculosis drug resistance.