It is also true that there can be variation in bias between 'technical replicates', data sets created from the same sample using the same protocols. For example, the HiSeq 'Kapa' human data set (data set 14) was created from three libraries and sequenced in fourteen lanes on two flowcells, with no deliberate variation in protocol at any point. Yet when bias statistics are computed lane-by-lane, one sees substantial variation in bias between libraries, and between flowcells - although not between lanes from the same library and flowcell (Table 4). Most notable is the between-flowcell variation of the G|C ≥ 80% motif, which is approximately two-fold undercovered in the first flowcell, but very well covered in the second. Possible sources of unexplained variation include variability of library construction instantiations, cluster amplification devices (cBot), flowcells, and HiSeq instruments that were used. Although variations between technical replicates are of interest, they are, for the most part, smaller than those observed between platforms.
