Deciphering the role of angiogenesis in glioblastoma: Integrative insights from transcriptomic profiling, single-cell sequencing and interpretable machine learning approaches.

Glioblastoma (GBM), a highly vascularized and aggressive primary brain tumor, presents unresolved questions regarding the functional significance of angiogenesis-related genes (ARGs) in disease progression. To systematically investigate this, we employed univariate Cox regression followed by LASSO regression analysis to identify prognosis-associated ARGs. These candidates were subsequently evaluated using six machine learning-derived survival prediction algorithms, with model interpretability achieved through SHapley Additive exPlanations (SHAP) analysis. Our studies revealed that six hub ARGs (BMP2, FSCN1, NET1, AEBP1, SEMA3G, and RAB37) were identified and incorporated into a novel risk stratification model (high and low risk groups). High-risk patients demonstrated significantly poorer overall survival in the CGGA-GBM, GSE43378, and GSE7696 cohorts. Moreover, the high-risk group exhibited an immunosuppressive tumor microenvironment profile (e.g., elevated Treg infiltration, P < 0.0001), increased extracellular matrix stiffness (via mechanosensing markers) and differential drug sensitivity (OncoPredict analysis). Immunofluorescence confirmed co-localization of the hub protein FSCN1 with vascular (CD31⁺), stromal (α-SMA⁺), and mesenchymal (Vimentin⁺) compartments, suggesting its multifunctional role in GBM pathobiology. In summary, our research established the ARG-derived prognostic signature validated across independent GBM cohorts and revealed concomitant immune and mechanical niche dysregulation in high-risk patients. The rationale for combined angiogenesis/stroma/immune modulation strategies was provided, with FSCN1 proposed as the potential therapeutic target.