SynOmics: integrating multi-omics data through feature interaction networks.

Abstract

The integration of multi-omics data is essential for achieving a comprehensive understanding of molecular systems and enhancing the performance of predictive models in biomedical research. However, many existing models have limited capacity to capture cross-omics feature interactions, which hinders the depth of integration. In this study, we introduce SynOmics, a graph convolutional network framework designed to improve multi-omics integration by constructing omics networks in the feature space and modeling both within- and cross-omics dependencies. By incorporating both omics-specific networks and cross-omics bipartite networks, SynOmics enables simultaneous learning of intra-omics and inter-omics relationships. Unlike traditional approaches that rely on early or late integration strategies, SynOmics adopts a parallel learning strategy to process feature-level interactions at each layer of the model. Experimental results demonstrate that SynOmics consistently outperforms state-of-the-art multi-omics integration methods across a range of biomedical classification tasks, highlighting its potential for biomarker discovery and clinical applications.