Fig. 1

A schematic illustration of the deSALT approach. a There is a gene (termed “Gene A”) with four exons (respectively marked by the colors blue, red, yellow, and green; introns are marked by gray color) and four reads that all sequence throughout the whole transcript (assuming that Gene A has only one isoform). Moreover, each of the reads has some sequencing errors (marked by the short black bars in the reads). b Alignment skeleton generation (first-pass alignment): for each of the reads (read1 is employed as an example), deSALT finds the MBs between it and the reference genome (marked as colored bars) and connects them to build an optimized alignment skeleton using an SDP approach. c Exon inference: deSALT integrates all of the generated alignment skeletons by mapping their involved MBs to the reference genome. The projections of the MBs are analyzed to infer exon regions in the reference genome. d–e Refined alignment (second-pass alignment): for each of the reads, deSALT finds additional local matches on the exons between or near the exons involved in the alignment skeleton. Further, it recognizes all the inferred exons related to the alignment skeleton or the newly found local matches as “hit exons” and stitches all of them to generate an LSRS. (The figure shows that there are two newly found matches on exon 2, and they help to recuse this exon to build a correct LSRS.) The read is then aligned with the LSRS to produce a refined alignment