A Network-of-Networks Framework for Multi-Omics Data Integration and Analysis via Graph Neural Networks

The integration of heterogeneous multi-omics data — spanning genomics, transcriptomics,proteomics, epigenomics, and metabolomics — remains one of the central open challenges incomputational biology. Existing approaches either flatten omics layers into feature matrices,losing relational structure, or adopt multilayer network formalisms that treat layers asindependent graphs coupled only by alignment edges. In this position paper we propose afundamentally different data model: a Network of Networks (NoN), in which each node of atop-levelgraphisitselfacompletegraph, definedrecursively. Thisrecursivestructurenaturallyencodes the hierarchical organisation of biological systems — from molecular interactionswithin an omics layer, through pathway-level modules, up to patient-level similarity networks— without collapsing any level of resolution. We formalise the NoN model with a rigorousrecursive graph definition, describe a bioinformatics infrastructure built on top of it, andoutline how heterogeneous Graph Neural Networks (GNNs) can operate across all levels ofthe hierarchy simultaneously. We argue that the NoN paradigm offers a principled, scalable,and biologically interpretable foundation for next-generation multi-omics analysis platforms,and we identify key research directions and open challenges that must be addressed to realisethis vision.