Glioblastoma (GBM), a highly heterogeneous and aggressive brain tumor, presents significant clinical challenges due to its frequent recurrence and poor prognosis. In this study, we employed high-dimensional weighted gene co-expression network analysis (hd-WGCNA) and single-cell transcriptomic analysis to investigate the molecular heterogeneity of GBM. We identified functional gene modules associated with tumor cell subpopulations exhibiting highly malignant traits, particularly linked to proteasome dysregulation. Intercellular communication analysis revealed extensive interactions between malignant tumor subpopulations and tumor microenvironment (TME), highlighting critical crosstalk with tumor-associated macrophages (TAMs) and T cells. Using machine learning, we developed risk scores based on these malignant gene modules, which effectively stratify GBM patients by prognosis and treatment response, particularly in relation to immunotherapy. Furthermore, we systematically evaluated pathway enrichment, genomic variations, and drug response differences across risk groups. Finally, we validated the oncogenic role of PSMC2, a key gene in the proteasome complex, demonstrating its role in promoting GBM progression through cell proliferation, invasion, and epithelial-mesenchymal transition (EMT). Our findings provide novel insights into GBM heterogeneity, prognosis, and therapeutic strategies, suggesting PSMC2 as a potential therapeutic target.