P078

1:Research Center Borstel - Leibniz Lung Center; 2:The German Center for Infection Research (DZIF)

Tuberculosis (TB) remains a major global health challenge, particularly due to the rise of multidrug-resistant strains of the Mycobacterium tuberculosis complex (MTBC). Mixed infections, where a patient is co-infected with genetically distinct MTBC strains, are increasingly recognized as contributors to treatment failure and misleading results in transmission analyses. However, their true prevalence is likely underestimated due to the limitations of standard genotyping techniques. We developed mixDetector, a novel bioinformatic tool based on the DBSCAN clustering algorithm, designed to detect mixed MTBC infections directly from whole-genome sequencing (WGS) data. The method was validated using synthetic mixtures of known strain combinations and in-silico simulated datasets, then applied to three large TB datasets from Hamburg (Germany), Samara (Russia), and KwaZulu-Natal (South Africa). mixDetector reliably identified mixed infections, including those between closely related strains or with different drug resistance profiles. In the Samara dataset (n = 1,206), 83 samples (6.9%) showed clear evidence of mixed infection. The method achieved high sensitivity (1.0) and specificity (0.97) at a 5% minor allele frequency threshold. Variants below this frequency could not be reliably detected, highlighting limitations in current variant-calling tools. Our findings demonstrate that mixDetector provides a robust and scalable solution for detecting MTBC mixed infections from WGS data without the need for labor-intensive subculturing. This represents a significant advancement for TB diagnostics, treatment strategies, and molecular epidemiology, particularly in high-burden and MDR-endemic settings.