Fig. 2: Reverse engineering of splicing factor networks to estimate splicing factor activities with VIPER.

a Estimation of splicing factor activity from exon inclusion signatures. The algorithm requires two inputs: (i) splicing signature, the difference between exon inclusion in a condition of interest (Cond. A) and a reference condition (Cond. B); (ii) splicing factor (SF) network, the splicing factor→exon interactions and their corresponding likelihood and mode of regulation (MoR). Splicing factor networks can be obtained using either empirical–from splicing factor experimental perturbations– or computational (ARACNe and Multivariate Linear Regression (MLR)) methods. We estimated activity using normalized enrichment scores (NES) from gene set enrichment analysis (GSEA), Pearson and Spearman correlation coefficients, and VIPER. b Benchmark outline for splicing factor activity estimation algorithms. Five benchmark datasets were generated from single-splicing factor perturbation experiments. Performance was assessed using ROC-AUC by comparing predicted activities to the known perturbed splicing factor, evaluated across perturbations (horizontal) and across splicing factors (vertical). c Evaluation of splicing factor activity estimation with different methods for activity estimation from an exon inclusion signature and a splicing factor→exon network. Computational networks were reverse-engineered using three different datasets of increasing size. For “Empirical” networks, performance reflects evaluation on a held-out benchmark dataset, with networks constructed using all others. Only when ENA is held out can empirical network performance be evaluated across truly independent cellular contexts. See Supplementary Fig. 2b for this specific evaluation. d Distribution of number splicing factors per target exon (left), and number of target exons per splicing factor (right) in empirical splicing factor networks. e Distributions of the number of targets (exons) per regulator (splicing factor) considering either all splicing factors categorized by “Core” splicing factors appear in Spliceosome Database⁴³, “RBP” splicing factors that appear in Gene Ontology⁸² category “RNA binding protein”, or “Other” splicing factors. f Evaluation of splicing factor activity estimation for each splicing factor in benchmark datasets using empirical splicing factor networks. In box and whisker plots in (c–f), the median is indicated by a horizontal line, boxes represent the first and third quartiles, whiskers extend to 1.5 times the interquartile range, and outliers are shown individually. Source data are provided as a Source Data file.