Fig. 3

Statistical interactions prioritized by binomialRF in TCGA cancers recapitulate known cancer driver genes. The statistical interaction gene networks (Top) indicate the pairwise biomarker interactions identified by the binomialRF algorithm for the breast (Left) and kidney (Right) cancer datasets. Key features are involved in multiple interactors (super-interactors; e.g., SPRY2; COL10A1). Features names (gene products) found in the literature as associated to cancer pathophysiology are shown in black; those also documented as driving cancer genes in COSMIC are shown in green (Methods); the remainder are grey. Two exemplar statistical interactions (one per dataset) are circled and the log expression of their gene products and of their ratios are shown in the bottom panels. The distribution separation across tumor (green) and normal (orange) cases indicates a potential interaction between these two genes across the cohorts