Fig. 4

Cross-omics networks distinguishing high-grade from low-grade tumors. a Sub-networks consistently upregulated in high-grade (G4/5) compared to low-grade (G1) tumors across all three layers (CNA, mRNA, and protein). b Same as in a but downregulated genes. Functional annotation of the sub-networks in a and b with more than one node is given. All edges in a and b are supported by either experimental or database evidence (STRING evidence ≥ 0.348). c CNA, RNA, and protein FCs of Network Component 1 from a. Samples are ordered by grade group (top bar). t test results comparing Network Component 1 members against no change (i.e., 0) are shown for each molecular layer along with the average FC across Network Component 1 members (“effect size”). The black box marks the selected matching samples from patients with G4/5 and G3 tumor areas, i.e., tumor sample pairs from identical patients. Those areas exhibit weak but common amplifications of Network Component 1 members at the CNA and RNA layers. mRNA samples in gray were removed due to low RNA quality. Gray bars at the bottom show the grade group of the patients (low, intermediate, high) where the samples have (mainly) come from. d Kaplan-Meier curves for “altered” and “unaltered” samples, where “altered” is defined as an effect size greater or equal to the median effect. Results for three independent studies, TCGA (left), MSKCC (middle), and Aarhus (right) using the corresponding CNA data when available (first row) and mRNA data (second row). The Cox model P value corresponds to the P value of the variable of interest (i.e., average copy number change (CNA) or average z-score (mRNA) of Network Component 1) from the fitted Cox model after adjusting for patient age (when available, i.e., for TCGA and Aarhus)