HyperNetWalk: A Unified Framework for Personalized and Population-Level Cancer Driver Gene Identification via Multi-Network Hypergraph Diffusion

Abstract: Identifying cancer driver genes is crucial for understanding tumor biology and developing precision therapies. However, existing computational methods often rely on single biological networks or population-level mutation patterns, limiting their ability to identify patient-specific drivers and leverage the complementary information from multiple network types. Here, we present HyperNetWalk, a novel computational framework that integrates multiple biological networks and hypergraph diffusion to identify driver genes at both personalized and cohort levels. In the first stage, HyperNetWalk integrates protein-protein interaction networks, gene regulatory networks, and dynamic co-expression networks through sample-independent random walks on patient-specific subnetworks to capture topological importance and expression perturbation effects. In the second stage, it refines predictions through hypergraph-based random walks that leverage cross-sample information while preserving individual mutational contexts. Comprehensive evaluation on 12 TCGA cancer types demonstrates that HyperNetWalk achieves superior or competitive performance compared to state-of-the-art methods in both personalized and cohort-level predictions. Notably, HyperNetWalk successfully identifies known driver genes with high precision while revealing cancer type-specific drivers that reflect distinct biological mechanisms. Our framework provides a unified solution for personalized and population-based driver gene identification, offering valuable insights for precision oncology and therapeutic target discovery.