Fig. 2
Cancer cell transcriptional network (CCTN) characteristics and validation. a, b Node degree distributions. c, d Functional annotation of network genes with respect to their node degrees. a, c Regulator genes. b, d Target genes. e Median gene-specific correlations between predicted and originally measured gene expression levels of individual genes in 13 TCGA cancer cohorts for CCTN including only significant edges (pink), CCTN using all edges (blue), and for random networks with the same complexity as CCTN with significant edges (gray). CCTN with significant edges predicts gene expression levels significantly better than CCTN with all edges (p<6×10−169) and random networks (p<2.2×10−308, Wilcoxon test). CCTN with significant edges was used for all subsequent analyses. f Cumulative p value distributions correlating experimentally measured and computationally predicted single-gene perturbations pooling results from all 13 TCGA cancer cohorts. Forward: p values of correlations between computed impacts flowing from a perturbed regulator to its targets and the corresponding experimentally measured impacts. The forward model specifies the basic CCTN properties that were used to make impact predictions (one-sided correlation test quantifying for each single-gene perturbation if the observed correlation between predicted and measured impacts is significantly greater than zero). Reverse: p values of correlations between computed impacts flowing in the reverse direction from the responding targets to their perturbed regulator and experimentally measured forward impacts. Random: Baseline for non-significant enrichment of small p values. See ‘Results and discussion’ and ‘Methods’ for details of the forward and backward models. The forward model predicted responses of single-gene perturbations significantly better than the reverse model (p<0.015 for each cohort) and than randomly expected (p<2.1×10−23 for each cohort, one-sided Kolmogorov–Smirnov test). CCTN cancer cell transcriptional regulatory network, sig. significant, TCGA The Cancer Genome Atlas