Recurrent Event Analysis and Biomarker Discovery for Glioblastoma Using mRNA-seq Data

Glioblastoma multiforme (GBM) is an aggressive brain tumor with poor survival outcomes and a high recurrence rate. This study employs Cox Proportional Hazards (CoxPH) and DeepSurv models to predict GBM recurrence, facilitating proactive patient care. To optimize model performance, three feature selection models–SelectKBest (SKB), Recursive Feature Elimination (RFE), and Sequential Feature Selection (SFS)–were applied to mRNA-seq data, generating three independent feature sets. The SFS-derived feature set performed best with CoxPH, while SKB outperformed other methods with DeepSurv, achieving the study's highest testing concordance index (c-index) of 0.6818. The effectiveness of SFS with CoxPH and SKB with DeepSurv stemmed from how each method complemented the strengths and limitations of its paired model. Three biomarkers-ZNF248, UNC93B1, and ZNF22-were identified as predictive of recurrence across feature sets. This study highlights the role of feature selection in enhancing GBM prognosis and demonstrates its utility in biomarker discovery, paving the way for personalized treatment strategies.