Fig. 5.

Results of the simulation study based on gene expression data from the TCGA LGG cohort without inter-gene correlation and representative pathways from the MSigDB Hallmark collection. a The correlation in the gene expression data has been broken by randomly permuting the values for each gene. Each panel plots the predictive performance of the evaluated gene-level and pathway-level models for simulation studies that associated survival with one of four Hallmark pathways (Hallmark estrogen response late, Hallmark E2F targets, Hallmark TGF beta signaling and Hallmark MYC targets V2 respectively) selected to represent the four possible combinations of large or small pathway size and high or low average inter-gene correlation. In these plots, the Cox concordance index is plotted on the y-axis with the x-axis representing the standard deviation of the Gaussian noise added to the simulated survival times. The error bars represent the standard error over 20 replications. b Heatmaps that represent the lack of inter-gene correlation for the four corresponding Hallmark pathways after random permutation of the gene expression values