Fig. 2

Derivation and validation of LungNet. a Using the multi-tissue RNA-seq compendium dataset from GTEX encompassing genome-wide gene expression measurements for > 8000 samples encompassing 30 tissue types, we inferred a lung-specific regulatory network for 38 TFs highly expressed in lung and a total of 1145 downstream gene targets. b Boxplot of TF-activity levels inferred using LungNet for each tissue-type in the same GTEX data, confirming the validity of the TF-activity estimation procedure. c Validation of LungNet in an independent multi-tissue RNA-seq dataset (NormalAtlas). Color bars compare the estimated average TF-activity levels of the 38 TFs between lung and all other 31 tissue types. In bold, we indicate those TFs which exhibit statistically significant higher TF-activity levels in lung. d Example boxplots of estimated TF-activity levels for five selected lung-specific TFs. P values are from a one-tailed Wilcoxon rank sum test. e Boxplot comparing t-statistics of differential TF activity between lung and all other tissues for the 38 TFs against the corresponding t-statistics obtained after randomizing the gene targets for each of the 38 TFs. P value is from a paired Wilcoxon rank sum test. f Scatterplot of t-statistics of differential TF activity (y-axis) against the t-statistics of differential TF expression (x-axis). Green dashed lines indicate significance threshold P = 0.05 for significantly positive statistics (i.e. higher activity or expression in lung tissue compared to all other tissue types). g Comparison of SEPIRA to simple differential expression (DE) analysis in predicting increased activity of the 38 LungNet TFs in the normal lung tissue of three independent gene expression datasets compared to other normal tissue types: the RNA-seq set from the ProteinAtlas (PrAtlas) and two microarray expression sets (Roth et al. and Su et al., see “Methods”)