An iterative and automated computational pipeline for untargeted strain-level identification using MS/MS spectra from pathogenic samples

Mathias Kuhring; Joerg Doellinger; Andreas Nitsche; Thilo Muth; Bernhard Y. Renard

The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. . https://doi.org/10.1101/812313 doi: bioRxiv preprint incorrect species. For some samples, correct strains are observed as a top candidate even in the original counts, independently of database choice and prior to count adjustments. Nevertheless, all samples benefit either from the iterative and focused database usage, the count adjustment procedure or the reduced resource consumption while the outcome remains legitimate. It was shown that TaxIt improves taxonomic assignment already in the count-based ranking, independently of count adjustments, by limiting candidates to strains of a single species early on ( Figure 3) . For instance, the search of bronchitis samples against the NCBI Blast NR bacteria subspace results in a rather uniform distribution of original counts for Infectious bronchitis virus strains: the strains Beaudette, Beaudette CK and Beaudette UK are only slightly increased in comparison to other strains (Figure 3B, Original) . In contrast, the iterative approach results in less strain candidates of the same species focusing solely on Beaudette strains when selectively searching against Avian coronavirus strains in the secondary iteration (Figure 3A, Original) . This is a consequence of how and whether parental or multispecies proteins are associated with strains or taxa in general within the NCBI taxonomy. In case of the iterative approach, fewer mutual species or genus proteins are consulted in the strain identification iteration. However, the extent of this effect varies between samples, candidate strains or target databases and taxonomies, respectively. For instance, bacterial strain proteomes such as the Bacillus subtilis strains feature numerous directly assigned mutual protein sequences and thus result in an extended range of candidate strains of the same species (Supplementary item 1 -Figure S3 and S4) . Nevertheless, the restriction to a specific set of strain proteomes prevents manifold primary misassignments to distant species, genera or even phyla as can be observed for Pipasic-and unique-PSMsbased original counts, respectively, that cannot be sufficiently resolved after correction (Supplementary item 1 - Figure S3 and S4) . In general, the iterative and selective database usage ensures that final strain selection is limited to strain candidates of an appropriate species. Thus, it prevents false positive hits on distinct strains of other taxa including species, genera and phyla and allows for a more confident final strain candidate selection. Furthermore, uniform distributions and even consensus in original counts of strain candidates demonstrate the need and benefit of count adjustment methods. The implemented weighting procedure can resolve ties between strains such as in the TaxIt cowpox sample analysis (Figure 2 ) or at least amplifies the correct strain and increases the distance to competing candidates. TaxIt infers exactly one strain for the presented samples each expect for the bronchitis sample where the strains Beaudette CK and Beaudette UK cannot be differentiated. We observed that the corresponding PSMs are fully shared between the two proteomes. Although different proteins are available for each strain in general, peptide hits are either assigned to shared proteins of the parent strain Infectious bronchitis virus or to homologous proteins that differ only in identifier but not in sequence. Though a more g