BIOM-49. PATIENT-CENTRIC INTEGRATED GRAPH DATABASE REVEALS CRITICAL BIOMARKERS IN THE RECURRENCE OF IDH WILD-TYPE GLIOMA

Abstract BACKGROUND IDH wild-type glioblastoma (GBM) represents a formidable therapeutic challenge due to its consistent recurrence and progression despite standard treatments. Understanding the molecular and cellular mechanisms driving GBM recurrence through data integration is crucial. This study employs a novel graph database strategy to uncover insights into these tumors and identify functional biomarkers through graph algorithms-based pathway enrichment analysis. METHOD We developed a stand-alone graph database using Neo4j for primary-recurrent paired GBMs, integrating clinical and gene expression data from the Glioma Longitudinal AnalySiS (GLASS) Consortium and internal MD Anderson cohort. Based on log2 fold-change, we used hierarchical trees for differentially expressed genes to decompose the expression profile. Nodes in the database represent patients with considerable absolute log2 values determined by hierarchical trees, while shared patients define the edges. We applied graph algorithms such as degree centrality and community detection to identify potential biomarkers and supplemented these analyses with pathway identification to refine the biomarker list. Further, we divided the data set into 80% training and 20% testing sets. We evaluated the predictive performance of these biomarkers using machine learning models, including logistic regression, random forest, SVM, and neural networks. RESULTS DEG analysis revealed 33 down-regulated and 418 up-regulated genes in recurrent tumors. The top enriched pathways included neuroactive ligand-receptor interaction and GPCR ligand binding. Our database approach implicated 35 out of 451 genes as significant drivers of recurrence in IDH wild-type GBM. The prediction performance of these biomarkers was comparable to features selected through methods such as mRMR and logistic lasso and target identification methods like WGCNA. CONCLUSION Our innovative systems biology approach reveals that ligand-receptor interactions are crucial in driving recurrence in IDH wild-type GBM. This method also yielded promising mechanistic targets for recurrence, prompting focused and therapeutically relevant functional analyses. Moreover, this methodology is not limited to our study but can be applied in other multi-omics settings, offering a robust tool for biomarker discovery and validation.