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

Besides identification efficiency, computational performance may delimit the potential application as routine or expeditious method in research or clinics. Therefore, we compared runtime and memory consumption for all samples and strategies (as illustrated in Table 2 and Figure 5) . Analysis was performed with X!Tandem as database search engine and limited to 24 threads on a server with Debian GNU/Linux 8.9 (jessie), 64 cores (128 threads) of type Intel(R) Xeon(R) CPU E5-4667 v4 @ 2.20 GHz, 512 GB of RAM and SSD storage. Applying the iterative approach reduces memory usage down to one third for viral strain identification and two third for bacterial strain identification. While the runtime of Pipasic is comparably high, TaxIt shows no substantial change in runtime in comparison to the unique-PSMs-based strategy for small databases such as the collective viral sequences or small sample sizes. However, analyzing the full bacillus sample reveals an increased runtime when using NCBI RefSeq proteins plus selected strain proteomes instead of the extensive NCBI Blast NR database.