New algorithm improves detection of bedaquiline resistance in tuberculosis

Bedaquiline (BDQ) has become a crucial drug for treating multi-drug resistant tuberculosis, but rising resistance threatens its effectiveness. Building reliable catalogues of genetic mutations linked to bedaquiline resistance is essential for diagnosis and public-health surveillance. Historically, creating those catalogues has demanded deep expert knowledge, complex grading rules, and procedures that are difficult or impossible to reproduce. To address this problem, a team led by Philip W. Fowler developed an automated, reproducible approach called catomatic. Rather than rely on bespoke expert judgement, catomatic applies clear statistical rules to associate genetic variants with resistance or susceptibility. The researchers used a large dataset to test the method and to examine how to classify different kinds of mutations, including variants present as minor subpopulations within a bacterial sample. Their work aimed to produce a publicly accessible, deterministic catalogue of resistance-associated variants that can be applied consistently in clinical and surveillance settings, removing the opaque steps that have made earlier catalogues hard to reproduce or audit.

The catomatic method links variants to phenotype using a two-tailed binomial test with a stated background rate and was applied to 11,867 Mycobacterium tuberculosis samples with whole genome and bedaquline susceptibility testing data. The team investigated how to group variants and evaluated the phenotypic importance of minor alleles that occur in subpopulations. They found that the genes mmpS5 and mmpL5 are not directly associated with bedaquline resistance. By focusing on variants in Rv0678, atpE, and pepQ, the resulting AMR catalogue achieved a cross-validated sensitivity and specificity of 79.4 ± 1.8 % and 98.5 ± 0.3%, respectively, for 94 ± 0.4% of samples. Importantly, identifying samples with subpopulations containing Rv0678 variants improved sensitivity, which led the authors to recommend lowering detection thresholds in bioinformatic pipelines. By using a more permissive and deterministic algorithm trained on a sufficient number of resistant samples, they reproducibly built a comprehensive and accurate catalogue of BDQ resistance-associated variants.

This work matters because it offers a practical, transparent path from genome data to actionable resistance calls. Bedaquiline is now globally endorsed for tuberculosis treatment, and clearer, reproducible genetic catalogues can help clinicians and public-health teams detect resistance earlier and more reliably. The catomatic approach simplifies the statistics behind variant classification and makes the process publicly accessible, enabling other researchers and laboratories to reproduce results and build on the catalogue. The findings about subpopulations and the role of Rv0678, atpE, and pepQ focus attention on where diagnostic pipelines should be most sensitive, and the conclusion that mmpS5 and mmpL5 are not direct markers of resistance helps avoid false associations. Overall, the study highlights the value of reproducible statistics and sustainably developed software in genetics-focused microbiology, and provides a tool and evidence base that can improve both clinical decision-making and surveillance for BDQ resistance.