Fig. 3

Experimental validation of differentially splicing predictions by SUPPA2. a Comparison of predicted and experimentally validated ΔPSI values for 83 cassette events differentially spliced between the double knockdown of TRA2A and TRA2B and control in MDA-MB-231 cells. We show the cumulative proportion of cases (y-axis) according to the absolute difference between the predicted and the experimental value (|ΔPSI − RTPCR|), for the events detected by each method: SUPPA2 (66), rMATS (78), and MAJIQ (72). Additionally, we give for each method the Pearson correlation R between predicted and experimental values. b False positive rate (FPR) calculated using 44 RT-PCR negative events. FPR was calculated as the proportion of the detected events that was found as significant by each method: SUPPA2 (1/31), rMATS (2/35), MAJIQ (2/36), DEXSeq(2/25). c Experimental validation by RT-PCR of a subset of novel events with TRA2B CLIP tags and Tra2 motifs. These events include cases that were only predicted by SUPPA2 (CHRAC1, NDRG3, METTL10) and cases that were not predicted by any method but were significant according to SUPPA2 before multiple test correction (ERLEC1, PYGL, DCAF10, HAUS8, EML4, UBA3) (Additional file 2: Table S14). RT-PCR validation was performed in triplicate. Error bars indicate the standard error of the mean. Cases that change significantly (p < 0.05) according to a two-tailed t-test comparing the three values of the knockdown versus control are indicated with an asterisk. d Experimental validation of a new skipping event in EML4 upon knockdown of TRA2A and TRA2B (three biological replicates shown in each case)