For each Ion Torrent quality score, we evaluated the empirical rate of error for bases assigned to that quality score. Consistent with previous reports [1], [2], we observe that the PGM quality scores underestimate the base accuracy, but observe that they have become more accurate with sequential sequencing kit releases ( Figure 9 ). The relationship between the empirical quality and the estimated quality is not strictly linear, potentially a consequence of the quality score look-up table method. As with Roche 454 quality scores, the qualities can be only used to detect inserted and substituted bases [11]. We evaluated whether there was any relationship between position within the homopolymer and the assigned confidence. We found that even in correctly called bases within a homopolymer, there is a decrease in assigned quality along the homopolymer length (Figure S10), as was found in Roche 454 pyrosequencing [11], although, overall the quality scores decrease more rapidly for inserted (error) bases. Counter intuitively, it is possible for the first base in a homopolymer run to have lower quality score than later bases, irrespective of
