Figure 2.

An example of biased loss of information encountered in our data when mapping to an existing reference. The reads contain a 30 bp insertion relative to the reference. Correct alignment, shown in the upper panel, would have inserted a 30 bp gap into the reference to accommodate this. What the mapper actually did (lower panel) was to align part of each read correctly either to the left of the insertion or to the right of it, and discard the rest of the read. ‘Read 1’ and ‘Read 2’ each represent roughly 2,000 similar reads; their consensus is therefore well supported but misses the insertion. This bias occurred despite the reference having been identified as the closest of 3,249 to this set of reads. Similar errors were made by the mapper’s smalt, BWA, and bowtie, resulting in the same erroneous consensus being called in each case. Bases in the reads that differ from the reference are shown in blue; the ends of the reads that were discarded during mapping (i.e. not aligned) are shown in grey with strikethrough. This figure corresponds to Position 8450 in Fig. 5.