The worldwide spread of antimicrobial resistance (AMR) bacteria is a serious global issue. Among them, Tuberculosis (TB) caused by Mycobacterium tuberculosis is the major global health problem caused from a single pathogen which listed in the 10 causes of death from 2000 to 2016. Constructing TB antidrug resistance prediction model is crucial because drug susceptibility testing in lab is time-consuming. Furthermore, it could help doctors in early diagnosis. However, current prediction sensitivity and specificity of Isoniazid(INH), Rifampicin(RMP) and Ethambutol(EMB) TB drugs are not ideal. To address the need, we utilize protein information of each assembled Mycobacterium tuberculosis genome in PATRIC database to develop the state-of-the-art drug resistance model with machine learning approach. Our results show that with enough amount of data, the trained model with specific hyperparameters can outperform the previous studies.
EE 4057 final project report
These authors contributed equally to the work