Fig. 3

The ζ and comparison with several key existing statistical approaches. a At each Z-score bin over a full Z-score range, the level of hits (expressed as the percentage of induced AS events over the total number of AS events monitored) is plotted with 10 representative splicing regulators (individually colored) compared to 10 non-expressors (grey). Left and right separately plot induced exon inclusion and skipping events. b At each Z-score bin over a full Z-score range, the level of hits in response to siPTBP1 (purple) or negative controls (NS-mix, green). An optimal SVM curve (black) is derived to maximally distinguish between true positives (siPTBP1) and true-negatives (NS-mix). c Calculation of a weighted ζ-score based on the area between the specific Z-score line of a gene (black) and the SVM curve (red). At each Z-score bin, the area is calculated by multiplying the Z-score, thus giving increasingly weights (purple) to hits at higher Z-scores. d The distribution of weighted ζ-score for annotated core spliceosome components among top 350 high-ranking genes. The top 10 high-ranking genes are enlarged (top). Only DEFB131A doesn’t belong to core spliceosome, which was later determined to result from off-targeting to SF3B1 (see Additional file 1: Fig. S4d). e, f The ROC (e) and PRC (f) curves are deduced using different software. Weighted ζ-score in two directions calculated by ZetaSuite are combined in this analysis to reflect the overall functional consequence. This is not applicable to other software, and we thus display the data separately. g The summary of the areas under all deduced ROC and PRC curves using different software