Comparison of classical multi-locus sequence typing software for next-generation sequencing data

Abstract

Multi-locus sequence typing (MLST) is a widely used method for categorizing bacteria. Increasingly, MLST is being performed using next-generation sequencing (NGS) data by reference laboratories and for clinical diagnostics. Many software applications have been developed to calculate sequence types from NGS data; however, there has been no comprehensive review to date on these methods. We have compared eight of these applications against real and simulated data, and present results on: (1) the accuracy of each method against traditional typing methods, (2) the performance on real outbreak datasets, (3) the impact of contamination and varying depth of coverage, and (4) the computational resource requirements.

Keywords: MLST; multi-locus sequence typing; next-generation sequencing; software comparison.

Fig. 1.

Number of correct calls of each application as coverage increases. Each ST consists of seven alleles, and all seven must be correctly and confidently called to calculate a ST.

Fig. 2.

Running time (s) of each application as the coverage increases to assess the impact of the depth of coverage. No assembled contiguous sequences could be generated where the coverage was less than 7×, as such no data was recorded for the reliant methods (mlst and MLSTcheck). No performance results are available for BioNumerics or SeqSphere+.

Fig. 3.

Disk space requirements in bytes for each software application as the depth of coverage increases. Due to the large difference between applications, a log scale is used.

Fig. 4.

STs called by each software application when given data containing two different Salmonella samples in varying ratios of abundance. Where there is no ST called, or where the ST has any ambiguity at all, it is marked as low confidence. A false positive is where an ST is called with high confidence and is not one of the two samples in the raw data.