Integrated Multi-Omics and Radiogenomic Analysis Identifies S100B and ITGB5 as Complementary Candidate Biomarkers of Glioblastoma Heterogeneity.

Glioblastoma (GBM) is characterized by marked intratumoral heterogeneity, diffuse invasion, and a profoundly immunosuppressive tumor microenvironment. In this study, we applied an integrative multi-omics and radiogenomic framework to identify candidate biomarkers reflecting complementary dimensions of GBM heterogeneity. Public bulk transcriptomic datasets were integrated to define GBM-related differentially expressed genes, which were intersected with curated microbiota-associated gene sets as a hypothesis-generating screening strategy. Machine-learning models with SHAP interpretation were used for candidate-gene prioritization, followed by survival-association analysis, pan-cancer comparison, Human Protein Atlas immunohistochemistry, CPTAC proteomics, immune infiltration analysis, single-cell transcriptomics, spatial transcriptomics, in silico perturbation analysis, and MRI-based radiogenomics. GBM samples showed enrichment of extracellular matrix organization, proliferative programs, immune-related signaling, and vascular or endothelial pathways, with relative reductions in neural, synaptic, and myelin-associated signatures. S100B and ITGB5 emerged as survival-associated candidate markers with different biological contexts. Multi-omics analyses suggested that S100B may reflect broadly distributed glial-lineage and malignant-state programs, whereas ITGB5 was more closely associated with focal extracellular matrix remodeling, stromal-vascular interactions, and immunoregulatory niches. Radiogenomic analysis further suggested distinct MRI-derived associations for the two genes, with the ITGB5-related model retaining a broader radiomic signature and showing a more heterogeneous habitat pattern than the S100B-related model. These exploratory findings support S100B and ITGB5 as complementary candidate biomarkers of GBM heterogeneity and provide a basis for future experimental, multicenter, and prospective validation.