YouTube video by Craig's Canvases

The initial ‘alignment’ stage of TRACS uses the hash-based search algorithm ‘Sourmash’ to identify a set of reference genomes that best represent the species and their strains in a sample14. Unlike alternative approaches, TRACS avoids the error-prone deconvolution of reads into different genome bins and instead aligns the entire read set to each reference genome individually, generating a count of the alleles observed at each site15. This enables the incremental addition of new samples and the integration of additional species into the reference database without requiring the reprocessing existing data. A set of statistical filtering algorithms are then applied to exclude regions affected by shared sequence homology, multi-mapping reads, poor alignment quality and low sequencing coverage (Fig. 1). This includes a scan statistic, similar to those used in phylogenetics to detect recombination16, that can identify regions with elevated polymorphism rates, often resulting from shared sequence homology between co-colonizing strains or gene duplications. TRACS also includes an empirical Bayes approach to account for regions of a reference genome with insufficient coverage to accurately represent multiple strains present within a sample (Methods). The resulting filtered alignments are converted into reference-based multiple sequence alignments (MSAs) for each reference genome. Finally, TRACS incorporates a...