Integrative Multi-Omics Profiling and Machine Learning Identify Key Molecular Determinants Distinguishing Glioblastoma from Lower-Grade Glioma

Background: Glioblastoma (GBM) and lower-grade glioma (LGG) are distinct glioma subtypes with divergent molecular landscapes and clinical outcomes. Accurate molecular characterization is critical for prognostic assessment and therapeutic decision-making. Methods: We performed an integrative multi-omics analysis combining RNA-sequencing, copy number variation (CNV), and somatic mutation data from TCGA. Feature selection across omics layers was applied to identify robust biomarkers, which were subsequently used in Random Forest and Logistic Regression classifiers for GBM–LGG discrimination. Cox proportional hazards regression and pathway enrichment analyses were conducted to assess prognostic relevance and functional context. Results: We observed pervasive CNV gains and losses in GBM relative to LGG, and subtype-specific mutation enrichment in canonical gliomaassociated genes including IDH1, ATRX, EGFR, PTEN, and TP53. Transcriptomic profiling revealed upregulation of RETN, LBX1, and CEACAM8 in GBM. Six overlapping genes (TP53, PTEN, RB1, RETN, LBX1, CEACAM8) spanned RNA, CNV, and mutation layers and served as top features in machine-learning models. Random Forest and Logistic Regression classifiers achieved high performance (test ROC–AUC = 0.965 and 0.975, respectively). Cox analysis indicated RETN and RB1 upregulation increased hazard, whereas PTEN and LBX1 CNV alterations were protective. Pathway enrichment highlighted cell cycle regulation, transcriptional control, senescence, and cancer-related signaling. Conclusion: Our integrative multi-omics analysis identifies a robust gene signature capable of accurate GBM–LGG classification and survival risk stratification, demonstrating the potential of combining bioinformatics and machine-learning approaches for precision diagnostics and prognostics in glioma.