Even after applying the quality trimming used for the initial Ion Torrent studies, we found that our estimate of quality-clipped accuracy for the 100 bp OneTouch kit was substantially lower than first reported. The first Ion Torrent study was conducted prior to the changes in the flow-cycle pattern, a change implemented to improve synchronicity between reads on the same bead. By analysing the flow-values, we are able to show that this change in the flow-pattern results in specific positions within the flow cycle being more susceptible to over-call/under-call error than others. Error-prone cycles tend to be ‘A’ or ‘T’ flows, and unlike Roche 454, over-calls of zero length homopolymers and under-calling homopolymers of length one are not improbable. We find that insertions are more common than deletions overall, however, under-calling (deletions) rapidly becomes the dominant error type with increasing homopolymer length. The variance of flow-values with respect to homopolymer length, cycle number, and position-within a cycle (PIC) will present challenges to studies, such as indel variant detection, which often assume a global base error rate. Consequently, higher read coverage and
