Can We Develop Glioma Subtype-Specific Precision Medicines? An Integrative Machine Learning Pipeline for Biomarker...

Glioma remains a major clinical challenge due to its molecular heterogeneity and limited therapeutic options. While numerous biomarker and drug discovery efforts exist, most are restricted by small sample sizes, subtype-agnostic analyses, or limited integration of computational strategies. Here, we present an integrative machine learning-based systems pipeline for the identification of subtype-specific biomarkers and repurposed therapeutics for glioblastoma (GBM) and low-grade glioma (LGG). We report high-confidence, subtype-specific biomarker candidates by harnessing publicly available gene expression datasets and systematic analyses with oversampling strategies to balance class distributions, followed by feature selection algorithms. Specifically, 10 candidate genes with strong diagnostic potential were identified, including RAB11FIP4, TYRO3, THEM5, SST, SMIM32, MIGA1, ARFGEF3, and ANK3 for GBM and GUCA1A and CES4A for LGG. Repurposed drug candidates were then predicted via signature-based prioritization and evaluated using molecular docking simulations, revealing six promising compounds for GBM (vandetanib, capecitabine, melatonin, agomelatine, ramelteon, and tasimelteon) and one for LGG (ambroxol). This study demonstrates the utility of combining class-balancing, feature selection, and drug repurposing pipelines to uncover clinically relevant glioma biomarkers and therapeutic candidates, thus providing a computational foundation for future experimental and translational validation in these brain cancers and neuro-oncology.