The occurrence of substitution errors in Roche 454 pyrosequencing has previously been attributed to PCR processes [16], with an expected substitution error every 105–106 bases [17]. While PCR does contribute to the substitution rate, the substitution rates observed in this study (see above), can not be explained by this alone. We found that a disproportionately large number of substitution errors (between 16%–33% across the datasets) had a low-confidence flow-value call of 0.51. This is in contrast to the percentage of ‘correct’ 1-mer calls with a 0.51 flow-value call (0.4%). All substitutions with the flow-value 0.51 had a quality score lower than 20, 99% of them had a quality of 12 or less, thus quality scores can be a reasonable indicator of PGM introduced substitution errors.
