Error models for Illumina v4, Illumina v5, and Roche/454 were analysed with GemStats. Error rates are summarised in Table 3. Striking differences between the error profiles of Illumina v4 and Illumina v5 are apparent, justifying the need for empirical chemistry- or run-specific error models when simulating NGS data. Combining results for the first and second reads in a pair, Illumina v5 had an error rate of 0.31%, five times lower than the error rate of Illumina v4 (1.66%). Illumina v4 also showed large differences between the first and second reads in a pair with second reads having considerably greater error rates. Roche/454 gave the least errors and its mismatch error rate of 0.12% is in line with recently published estimates [8]. Consideration of the top five mismatches within their sequence-context suggests that all runs had some trouble with homopolymers, particularly in G or C homopolymers for the Illumina runs. This problem has been well document for Roche/454 data [8,20], and there is some evidence that homopolymers are also problematic for Illumina data [21]. Our data also supports the known problem with the CGG motif in Illumina sequencing [9,21].
