Identification of SST as a glioblastoma diagnostic biomarker via integrated bulk and single-cell transcriptomics and DrugReflector AI-driven drug discovery

This study aims to identify reliable diagnostic biomarkers and explore their biological significance in glioblastoma (GBM). The bulk transcriptomic dataset GSE68848 was used for differential expression and weighted gene co-expression network (WGCNA) analyses, and core genes were screened by three machine learning algorithms (LASSO, SVM-RFE, and Random Forest) and cross-validated in the independent dataset GSE50161. Single-cell RNA sequencing (scRNA-seq) dataset GSE273274 was analyzed to determine the cellular localization of the candidate gene and to characterize the tumor microenvironment. Functional consequences were simulated by scTenifoldKnk in silico knockout, and candidate drugs were screened using the DrugReflector AI platform and evaluated by molecular docking. Cross-screening identified SST as the sole core feature gene. SST exhibited robust diagnostic performance in both the training set (AUC = 0.986) and the validation set (AUC = 0.846), with consistent downregulation confirmed by TCGA, CPTAC, and Human Protein Atlas data. Single-cell analysis showed that SST is specifically expressed in neurons, and its downregulation in GBM bulk tissue reflects neuronal loss in the tumor microenvironment. CIBERSORT analysis revealed that SST-high tumors displayed elevated proportions of activated NK cells and reduced M0/M1 macrophage infiltration. In silico knockout of SST perturbed 21 downstream genes enriched in neurotransmitter uptake and ion-transport-related pathways. DrugReflector ranked the retinoic acid receptor agonist TTNPB as the top candidate for reversing the GBM transcriptomic signature; however, molecular docking indicated that its binding to the SST receptor SSTR2 was nonspecific. SST serves as a robust diagnostic biomarker reflecting GBM-associated neural-microenvironment disruption, and its regulatory mechanisms and therapeutic potential warrant experimental validation.