Integrative multi-platform gene expression and machine learning analysis for glioblastoma biomarker discovery with experimental validation

Background: Glioblastoma (GBM) is a highly aggressive form of brain tumor with poor prognosis. This study aimed to identify genes critical to glioma development and assess their potential diagnostic and therapeutic value through an integrated analysis of microarray and RNA-Seq datasets. Methods: We used data integration to improve statistical robustness and reliability. LASSO regression, along with seven machine learning classifiers, was utilized to find the most important features. The classifier performance was validated on unseen datasets, where SVM and kNN showed the highest accuracy. Functional enrichment analysis was performed on the selected features to assess their biological relevance, followed by validation of biomarkers through qPCR. Results: We identified three key genes, PLOD1, COL4A2, and COL5A2, that emerged as consistently top-ranked across models. Survival analysis revealed PLOD1 is the only gene significantly associated with poor prognosis in GBM. Further validation in patient samples confirmed a strong association between high PLOD1 expression and GBM, highlighting its potential as a diagnostic biomarker and therapeutic target. Conclusion: This robust integrative machine learning approach for biomarker discovery in GBM supports the development of more personalized treatment strategies.