Fig. 3

An example of the alignments of simulated reads by various aligners This figure represents the snapshots of the alignments of the reads from the simulated 30× ONT 2D human dataset, around FCER1G gene (Chr1: 161215297–161219248) of GRCh38. FCER1G gene has 6 exons and 3 isoforms. According to the ground truth, there are 88 reads in this region. The numbers of Read100 and Read80 reads of deSALT are 84 and 88, respectively, higher than those of GMAP (#Read100: 45 and #Read80: 54), Minimap2 (#Read100: 19 and #Read80: 23), and Graphmap2 (#Read100: 50 and #Read80: 78). a The Sashimi plots represent the overall views of the alignments. Compared to the ground truth (the bottom track), it is observed that the deSALT alignments are more homogenous, i.e., at each splicing site, most of the reads have similar breakpoints, which also coincide with the ground truth. The more heterogeneous alignments of GMAP, Minimap2, and Graphmap2 are usually due to some less accurate alignments at small exons and exon boundaries. b A detailed view at the fourth exon of the FCER1G gene (length: 21 bp). deSALT correctly aligns all of the 88 reads spanning this exon; however, the corresponding numbers for GMAP (68), Minimap2 (24), and Graphmap2 (53) are lower. c A detailed view at the third exon of the FCER1G gene (length: 36 bp). It is observed that the reads have nearly the same breakpoints with the homogeneous alignments of deSALT. However, for the other three aligners, the breakpoints of the reads are more divergent, and some of them are less accurate, which could be due to the effect of sequencing errors as well as to the nearby small exons