Feature weighted models to address lineage dependency in drug-resistance prediction from Mycobacterium tuberculosis genome sequences.

MOTIVATION: Tuberculosis (TB) is caused by members of the Mycobacterium tuberculosis complex (MTBC), which has a strain- or lineage-based clonal population structure. The evolution of drug-resistance in the MTBC poses a threat to successful treatment and eradication of TB. Machine learning approaches are being increasingly adopted to predict drug-resistance and characterize underlying mutations from whole genome sequences. However, such approaches may not generalize well in clinical practice due to confounding from the population structure of the MTBC. RESULTS: To investigate how population structure affects machine learning prediction, we compared three different approaches to reduce lineage dependency in random forest (RF) models, including stratification, feature selection, and feature weighted models. All RF models achieved moderate-high performance (area under the ROC curve range: 0.60-0.98). First-line drugs had higher performance than second-line drugs, but it varied depending on the lineages in the training dataset. Lineage-specific models generally had higher sensitivity than global models which may be underpinned by strain-specific drug-resistance mutations or sampling effects. The application of feature weights and feature selection approaches reduced lineage dependency in the model and had comparable performance to unweighted RF models. AVAILABILITY AND IMPLEMENTATION: https://github.com/NinaMercedes/RF_lineages.

Large-scale statistical analysis of Mycobacterium tuberculosis genome sequences identifies compensatory mutations associated with multi-drug resistance.

Tuberculosis (TB), caused by Mycobacterium tuberculosis, has a significant impact on global health worldwide. The development of multi-drug resistant strains that are resistant to the first-line drugs isoniazid and rifampicin threatens public health security. Rifampicin and isoniazid resistance are largely underpinned by mutations in rpoB and katG respectively and are associated with fitness costs. Compensatory mutations are considered to alleviate these fitness costs and have been observed in rpoC/rpoA (rifampicin) and oxyR'-ahpC (isoniazid). We developed a framework (CompMut-TB) to detect compensatory mutations from whole genome sequences from a large dataset comprised of 18,396 M. tuberculosis samples. We performed association analysis (Fisher's exact tests) to identify pairs of mutations that are associated with drug-resistance, followed by mediation analysis to identify complementary or full mediators of drug-resistance. The analyses revealed several potential mutations in rpoC (N = 47), rpoA (N = 4), and oxyR'-ahpC (N = 7) that were considered either 'highly likely' or 'likely' to confer compensatory effects on drug-resistance, including mutations that have previously been reported and validated. Overall, we have developed the CompMut-TB framework which can assist with identifying compensatory mutations which is important for more precise genome-based profiling of drug-resistant TB strains and to further understanding of the evolutionary mechanisms that underpin drug-resistance.