(A) MAJIQ’s analysis pipeline. RNA-Seq reads are combined with an annotated transcriptome to create splice graphs and detect LSVs for each gene, then LSVs are quantified and compared between conditions. The visual output (VOILA) lists LSVs with violin plots representing estimates of percent inclusion index (PSI, Ψ) or changes in inclusion (dPSI, ΔΨ). Two cases are illustrated, for a single source three way LSV (orange), and a single target two way LSV (pink). (B) Correspondence between E[Ψ] by MAJIQ and Ψ by RT-PCR. R is the correlation coefficient. Colors and shapes represent different experimental conditions: mouse cerebellum and liver (dark and light orange diamonds, respectively); human unstimulated and stimulated T-Cells (dark and light purple dots, respectively). Total n = 208. (C) Correspondence between E[ΔΨ] by MAJIQ and ΔΨ by RT-PCR, where |ΔΨRT|>0.2. R is the correlation coefficient. Changes in inclusion were measured between liver and cerebellum mouse tissues (diamonds, n = 45); stimulated and unstimulated T-Cells (dots, n = 9). (D) Reproducibility ratio (RR) of high confidence differentially included LSVs, i.e. LSVs for which P(|ΔΨ|> 0.2) > 0.95), when comparing RNA-Seq from two conditions. A differentially included LSV is considered replicated if it maintains a rank at least as high as N in biological replicates, where N is the set size. LSVs are ranked by E[ΔΨ] and filtered for overlap. Twelve replicate pairs from Keane et al. (2011) were used to compute the histogram’s std (light blue). Other lines show MAJIQ’s RR with replicates (thick blue), RR for AS events detected by rMATS w/wo replicates (light and dark green), MISO (red), and RR for LSVs using Naïve Bootstrapping (orange). The inset bar chart shows the number of LSVs or AS events (N) derived by each method and used in the RR plots (see Materials and methods for more details).