Fig. 2

Increasing readout number leads to diminishing screen specificity with common statistical approaches. a, b The distribution of Z-scores based on 5 randomly selected alternative splicing (AS) events monitored in our screen (a) or all AS events measured (b) in response to siRNAs against 50 randomly selected non-expressed genes. The AS event was marked as red if the Z-score is >=3. c, d The Specificity based on common cutoffs (c, Z-score>=3) or SSMD value (d, SSMD value>=2) when different numbers of AS events were monitored. The specificity (defined by 1 minus the number of non-expressors scored as hits over the total number of non-expressors) is the mean value of 50 replicates under each condition. e Illustration of the principal theory to determine hits based on RSA, MAGeCK, and RIGER. Induced changes in AS are first ranked and the effects of knocking down a given gene on individual AS events are displayed as red bars. A hit would show enriched AS events in one direction (top) while a non-hit would display a relatively random distribution (bottom). f, g The distribution of induced AS events (based on Z-scores of induced exon skipping from left to right at top or induced exon inclusion from right to left at bottom) in response to knockdown SF3B1 (f) or SRSF2 (g). h The false discovery rate (FDR=FP/(FP+TP)) at different cutoffs with different methods. The FDRs at x-axis were calculated by different software (RSA, RIGER, MAGeCK, and CB²). The FDRs at y-axis were deduced based on the non-expressors and built-in positive controls (siPTBP1). False positive (FP): non-expressors; true positive (TP): siPTBP1-treated samples