Architecture and information flows of the IHAS inference machine.

The single cancer type data analysis (the top box) is undertaken for each cancer type separately. In each cancer type the TCGA omics data are processed and fed into the model inference algorithm to build Association Models and Association Modules. The Association Modules and mRNA expression data are bi-clustered to form Super Modules and Sample Groups. The (effector,target) pairs in the Association Models are used to construct the Artery Network from a unified biomolecular network. The Super Modules and Sample Groups then undergo subtype alignments, functional enrichment, and prognostic associations. The inference outcomes of individual cancer types are integrated to form pan-cancer subunits (Super Module Groups, Gene Groups, Consensus Artery Network). These pan-cancer structures are aligned with pan-cancer phenotypes (molecular phenotypes and prognosis), characterized (Recurrent Effectors, functional enrichment, and hubs in the Consensus Artery Network), and validated on external data (individual cancer types, perturbations, normal tissues). Black lines indicate TCGA data and association outcomes. Red lines indicate external data for validation. Blue lines indicate non-TCGA data used to infer IHAS.

(TIFF)