Fig. 2

a Comparison of the RATTLE similarity score (x axis), based on the longest increasing subsequence, with a similarity score calculated from Minimap2 (y-axis), using k = 6. Each dot represents a comparison between a pair of reads simulated from the same (orange) or different (blue) transcripts from two different genes. The distribution of values for each comparison is given as box plots along each axis. b Number of common bases between two reads (x-axis) and variance in the distribution of gap-length differences between adjacent matching k-mers between the same two reads (y-axis). Reads were simulated from two transcript isoforms from the same gene differing only by an internal exon (Additional file 1: Fig. S1e). Each dot is colored according to whether the reads originated from the same transcript, 1–1 (red) or 2–2 (blue) or not, 1–2 (green). c Clustering accuracy of RATTLE, CARNAC, and isONclust in terms of the V-measure (y-axis), using simulated reads. Simulations (x-axis) were performed with a single (SI) or with multiple (MI) transcript isoforms per gene, using a different number of total transcripts (t) and a different number of reads per transcript (r), indicated as SIt-r or MIt-r. Other accuracy metrics are provided in Additional file 2: Table S2. d Clustering accuracy using spike-in isoform RNA variants (SIRVs) as reference. The plot shows the V-measure (y-axis) for RATTLE, CARNAC, and isONclust using SIRV reads from four of the tested samples (x-axis): three using the cDNA-seq protocol (cDNA1, 2, 3) and one using the dRNA protocol (RNA1) (Methods)