Fig. 5

An example of the alignments of real sequencing reads by various aligners. This figure represents the snapshots of the alignments of the reads from the human ONT cDNA dataset around the VDAC3 gene (Chr8: 42391761–42405937) of reference GRCh38. The VDAC3 gene has 10 exons and 12 isoforms (according to Ensembl gene annotation). deSALT, GMAP, Minimap2, and Graphmap2 respectively mapped 2652, 2639, 2462, and 2480 reads to this region. The ratios #BaseGA/#BaseT of the aligners are respectively 78.93% (deSALT), 74.2% (GMAP), 72.69% (Minimap2), 74.02% (Graphmap2), where #BaseT is the total number of bases aligned to the VDAC3 region. This indicates that deSALT produces overall more accurate split alignments. Moreover, the #ReadGA statistics of the aligners are respectively 1630 (deSALT), 889 (GMAP), 751 (Minimap2), and 1030 (Graphmap2), also indicating that deSALT produces better full-length alignments. a The overall views (sashimi plots) of the alignments indicate that deSALT produces more homogenous alignments. Considering the higher #BaseGA/# BaseT and #ReadGA statistics, such alignments could be more plausible. b A detailed view at the second exon of the VDAC3 gene (exon length: 40 bp). deSALT aligns much more reads (i.e., 1703 reads) to this short exon than that of GMAP (1267 reads), Minimap2 (878 reads), and Graphmap2 (920 reads), indicating that deSALT potentially handles it better. c A detailed view at the 3′ splicing site of the fifth exon of the VDAC3 gene (exon length: 153 bp). It shows that the alignments of deSALT near the splicing site are more homogeneous, and the breakpoints of the reads coincide with the annotation, while the alignments of other approaches are more heterogeneous and seem less accurate