Tuberculosis is an infectious disease caused by Mycobacterium tuberculosis and remains a public health concern with 1.3 million deaths annually. Resistance to anti-tuberculosis drugs is recognized as a serious problem to global tuberculosis control. Pyrazinamide, a first-line drug, and bedaquiline, kept as a last- resort, are active against both actively replicating and dormant bacilli, which makes them an integral component of most drug cocktail regimens to treat tuberculosis. Despite their pivotal role in TB treatment, there are no WHO or FDA approved tests for drug susceptibility, with current approaches being unreliable, time-consuming and labor-intensive. Using the mCSM platform to understand the structural and functional consequences of resistance and susceptible mutations, we have developed an empirical pipeline to identify novel drug resistance for these drugs. The generated scores were trained using supervised machine learning algorithms to predict the susceptibility profile of each possible variant, which has been made available through user-friendly webservers SUSPECT-PZA and SUSPECT-BDQ. We found resistance mutations in pyrazinamide predominantly led to the destabilization of proteins involved in drug activation and altered binding to drug target in the case of bedaquiline. Working with the Victorian TB Program, we have implemented this approach to guide clinical patient treatment decisions. Therefore, we have shown for the first time that protein structure-guided interpretation of clinical genomic sequencing can accurately identify resistant infections. In the next phase of the project, we are trying to combine this structural information together with epidemiological models, to identify which mutations are more likely to arise in a population.