Integrative machine learning reveals hidden and emerging co-regulatory gene networks for multi-phase glioblastoma outcome prediction.

backgroundGlioblastoma (GBM) is a highly prevalent and aggressive type of brain tumor characterized by profound molecular complexity and poor prognosis. While conventional biomarker studies focus on highly significant genes or proteins associated with cancer outcomes, the contribution of gene-gene co-regulation to GBM progression remains unclear.

methodsThis study employs an application of integrative machine learning approach, utilizing a high-dimensional transcriptomic profile that considers gene-gene co-regulations to identify predictive gene networks involved in GBM occurrence and 1-year survival prediction. We further integrate these network models with both empirical protein-protein interaction (PPI) data and random walk-based information flow analysis across the PPI landscape.

resultsThis dual-layered approach uncovers gene modules that bridge transcriptional co-regulation with functional connectivity at the protein level. This integration highlighted several hub genes, including both strong and weak (e.g. BSN, RHOC, ANXA1, CSF1R, and ITGAM), that emerged as key molecular connectors involved in critical GBM processes such as immune response and neuronal signaling. Notably, these hub genes also exhibited cross-disease associations with traits including gut microbiome composition, type 2 diabetes, coronary artery disease, and other cancers, underscoring their systemic biological relevance.

conclusionOverall, our findings through the computational approach underscore the significance of co-regulatory gene networks in GBM biology. It also demonstrates how integrating transcriptomic and protein-level interactions can refine prognostic modeling, advance biomarker discovery, and inform future therapeutic development.