SplitStrains, a tool to identify and separate mixed Mycobacterium tuberculosis infections from WGS data

Abstract

The occurrence of multiple strains of a bacterial pathogen such as M. tuberculosis or C. difficile within a single human host, referred to as a mixed infection, has important implications for both healthcare and public health. However, methods for detecting it, and especially determining the proportion and identities of the underlying strains, from WGS (whole-genome sequencing) data, have been limited. In this paper we introduce SplitStrains, a novel method for addressing these challenges. Grounded in a rigorous statistical model, SplitStrains not only demonstrates superior performance in proportion estimation to other existing methods on both simulated as well as real M. tuberculosis data, but also successfully determines the identity of the underlying strains. We conclude that SplitStrains is a powerful addition to the existing toolkit of analytical methods for data coming from bacterial pathogens, and holds the promise of enabling previously inaccessible conclusions to be drawn in the realm of public health microbiology.

Anti-tuberculosis treatment strategies and drug development: challenges and priorities

“…The report provides a reference standard for the interpretation of mutations conferring resistance to all first-line and a variety of second-line drugs. It complements an elegant tool called SplitStrains , which leverages whole-genome sequencing data to identify and separate mixed M. tuberculosis infections with genetically different strains 90 , a phenomenon that can lead to hetero-resistance and the ensued treatment complications 91 , 92 . The new toolkit enables more precise detection, identification and quantification of multiple infecting strains within a sample.…”

Demixer: A probabilistic generative model to delineate different strains of a microbial species in a mixed infection sample

“…• Datasets LDAmix1 and LDAmix2 are synthetic datasets created using our SNP-LDA model's generative process to evaluate Demixer's potential in delineating the mixed strains under different hyperparameter combinations and deciding if a simple hyperparameter value would work well in further analyses. • To mimic the benchmark dataset used to evaluate QuantTB, we generated two standard datasets (ART-TBmix1 and ART-TBmix2) as per the simulation procedure described in [14][12] using ART simulator [33]. Both ART-TBmix1 and ART-TBmix2 consists of 800 samples at four distinct levels of coverage: 10x, 20x, 90x-10x, and 70x-30x, each with 200 samples.…”

“…Although this method is simple and does not require any overhead of keeping track of the new strains, it reveals only the strains' proportions and not their identity (mutational profile), and also cannot robustly handle cases when strains are present in equal proportions. SplitStrains [14] is a new statistical-based method that aligns with MixInfect due to its non-reliance on the reference database. The procedure uses a likelihood ratio test to resolve the heterogeneity of the samples, followed by the assignment of reads to a strain using the Expectation-Maximization (EM) algorithm and Naive Bayes classifier, in order to finally report the strain proportions.…”