Furthermore, the accuracy of PGM quality scores has improved markedly with each kit release. The addition of HRI-based clipping (as of Torrent Server 2.2.0) to complement the relatively lax quality trim has proven extremely effective at removing the error-prone 3′ end of the reads, albeit at a cost of 20–30 bp of read length. For datasets processed with older versions of Torrent Server, we highly recommend re-running the PGM analysis to improve run quality. However, trimming cannot remove all errors. Thus, we recommend that researchers who intend to use or develop methods for analysing PGM data take into account that the PGM has a higher error-rate than both Illumina and 454, and that PGM global base error rates are poor substitutes for flow error rates. Armed with the models developed in this study, bioinformaticians can develop platform-specific approaches for the PGM that adequately account for the majority of errors introduced by this platform.
