TBtypeR: Sensitive detection and sublineage classification of low-frequency Mycobacterium tuberculosis complex mixed infections

Mixed infections comprising multiple Mycobacterium tuberculosis Complex (MTBC) strains are observed in populations with high incidence rates of tuberculosis (TB), yet the difficulty to detect these via conventional diagnostic approaches has resulted in their contribution to TB epidemiology and treatment outcomes being vastly underrecognised. In endemic regions, detection of all component strains is crucial for accurate reconstruction of TB transmission dynamics. Currently available tools for detecting mixed infections from whole genome sequencing (WGS) data have insufficient sensitivity to detect low-frequency mixtures with less than 10% minor strain fraction, leading to a systematic underestimation of the frequency of mixed infection. Our R package, TBtypeR, identifies mixed infections from whole genome sequencing by comparing sample data to an expansive phylogenetic SNP panel of over 10,000 sites and 164 MTBC strains. A statistical likelihood is derived for putative strain mixtures based on the observed reference and alternative allele counts at each site under the binomial distribution. This provides robust and high-resolution sublineage classification for both single- and mixed-infections with as low as 1% minor strain frequency. Benchmarking with simulated in silico and in vitro mixture data demonstrates the superior performance of TBtypeR over existing tools, particularly in detecting low frequency mixtures. We apply TBtypeR to 5,000 MTBC WGS from a published dataset and find a 6-fold higher rate of mixed infection than existing methods. The TBtypeR R package and accompanying end-to-end Nextflow pipeline are available at github.com/bahlolab/TBtypeR.