Fig. 1

General work flow of CIDANE. Mandatory inputs (mapped RNA-seq reads and exon boundaries) and optional inputs (TSS, TES, and known transcripts) are used to summarize read alignments into segment covers, which count reads falling into non-ambiguously spliced segments of genes. From the corresponding splicing graph representation [37], an initial set of candidate isoforms is derived and a subset of expressed isoforms with estimated abundances is predicted by a regularized regression method during phase I. This set forms the input to the optional phase II, where improving isoforms are built on demand by a delayed column generation approach. New candidates inferred in phase II are then added to the initial candidate set to achieve a better fit of the model. After re-estimation of abundances and filtering (post-processing), a list of isoforms with abundance estimates is returned in gtf format