The majority of HFI errors were single-base insertions or deletions and only 29% of these occurred in homopolymeric regions, in contrast, 82% of strand-asymmetric HFIs occurred in a homopolymer of length two to three. In general, HFI errors manifested most commonly as insertions of A/T or deletions of C/G (Figure S3a). Strand-asymmetric HFIs were dominated by deletions of C and G (Figure S3b). There was no relationship between the base or flow positions of these errors across the reads with the same HFI. It will require further experimentation to identify whether HFIs are introduced by the library preparation, template preparation or the sequencer itself. The HFI error rate relative to the reference genome was approximately 1 in 1000 to 1 in 2000 bp, and HFIs accounted for 0.03%–0.09% of all bases in each dataset. Given that it is difficult to distinguish HFIs from bona fide polymorphisms without sequencing the same template over multiple chips, PGM sequencing may be compromised in polymorphism detection and amplicon sequencing projects. Here, we mask the HFIs from downstream analyses, as they are unlikely to be the consequence of individual flow-call inaccuracy, and will introduce bias into modeling of base and flow-call accuracy.
