Comprehensive multi-omics and machine learning framework for glioma subtyping and precision therapeutics.
Journal:
Scientific reports
Published Date:
Jul 10, 2025
Abstract
Glioma is a highly heterogeneous and aggressive brain tumour that demands an integrated understanding of its molecular and immunological landscape. We collected multi-omics data from 575 TCGA diffuse-glioma patients (156 IDH-wild-type WHO-grade 4 glioblastomas and 419 IDH-mutant WHO-grade 2/3 diffuse gliomas) together with two validation cohorts (CGGA n = 970; GEO n = 110). Using the MOVICS framework, we derived three integrative molecular subtypes-CS1, CS2 and CS3. Ten machine-learning algorithms in MIME were benchmarked, and the Lasso + SuperPC combination yielded an eight-gene GloMICS (Glioma Multi-Omics Consensus Signature) prognostic score. The subtypes display discrete biology: CS1 (astrocyte-like) is characterized by glial lineage features, immune-regulatory signaling, and relatively favorable prognosis; CS2 (basal-like/mesenchymal) shows epithelial-mesenchymal transition, stromal activation, and high immune infiltration, including PD-L1 expression; CS3 (proneural-like/IDH-mut metabolic) exhibits metabolic reprogramming (OXPHOS, hypoxia) and an immunologically cold tumour microenvironment (TME). CS2 is associated with the worst overall survival, whereas CS1 confers the most favourable outcome. Dual checkpoint blockade or T-cell-rejuvenation strategies may benefit CS2 tumours, while metabolic inhibitors could prove effective in CS3. The eight-gene GloMICS score outperformed 95 published prognostic models (C-index 0.74-0.66 across TCGA, CGGA and GEO). TME deconvolution, immune checkpoint profiling and TIDE analysis indicate that high-risk GloMICS tumours harbour immunosuppressive fibroblast-rich niches and exhausted CD8⁺ T cells. Connectivity-map screening nominated dabrafenib, irinotecan and three additional CTRP/PRISM compounds as candidate agents for the high-risk group. Our study establishes robust glioma subtypes and a transferable prognostic signature, offering a blueprint for biomarker-guided therapy. Future work should include single-cell and immunohistochemical validation of subtype hallmarks and prospective trials stratified by GloMICS score.
