Extended Data Fig. 5: OncoSigRF KRAS SigMaps exhibit tissue context specificity.
From: Oncoprotein-specific molecular interaction maps (SigMaps) for cancer network analyses
a, ROC curves for the OncoSigRF KRAS SigMaps in LUAD (red), LUSC (gray) COAD (brown), and PAAD (orange) for FPR ≤ 0.05. Performance is evaluated as the recovery of established KRAS pathway proteins. b, Gene set enrichment analysis (GSEA) of KRASMut synthetic lethal partners, as determined by Corcoran et al27 (N = 48, blue lines). To avoid training and testing on the same proteins, OncoSigRF predictions for COAD-specific KRAS SigMap proteins were obtained by training on a modified PGSS from which any established KRASMut synthetic lethal protein had been previously removed. Enrichment analysis was performed with the aREA (analytic Rank-based Enrichment Analysis) algorithm10. c, Scatterplot of OncoSigRF scores for KRAS SigMap proteins in PAAD-vs-LUAD (N = 19,789). Each dot represents the scores for one protein. Darker colored points have high scores (SRF ≥ 0.5) in at least one context, and lighter colored points score poorly in both contexts (SRF ≤ 0.5). R2PAAD/LUAD = 0.037. d, OncoSigRF COAD-specific KRAS SigMap in the form depicted conceptually in Fig. 1a. To prevent visual cluttering, only the top 33 OncoSigRF predictions (FPR ≤ 0.01) that are also VIPER-inferred KRAS interactors (p ≤ 0.01), PrePPI-predicted KRAS physical interactors, or both, are depicted. Bold and regular text node labels represent established and novel predictions, respectively; orange and blue node colors represent upstream regulators and downstream effectors, respectively; red, blue, and black node borders represent predictions that are druggable (Drug Repurposing Hub22), KRASMut synthetic lethal from the literature and validated here (see text), and both, respectively; orange and blue solid lines and gray nodes represent PrePPI-predicted physical interactors of KRAS.
