Abstract:Abstract BACKGROUND Glioblastomas (GBM) is a common primary malignant brain tumor. This study performed RNA-seq with 227 glioma samples with the aim to identify the immune-related gene markers for glioma prognosis prediction. METHODS Differentially expressed genes (DEGs) were identified between glioma and healthy samples in the TCGA dataset. Differentially expressed immune-related genes (DE-IRGs) were determined by intersecting DEGs to the immune-related genes in the ImmProt database. Independent survival associated DE-IRGs were selected by univariate and multivariate cox-regression analysis in TCGA dataset and validated in the CGGA databases and our RNA-seq data. Further enrichment analysis demonstrated the hub-genes among those prognosis-associated DE-IRGs. After the expression of hub-genes were confirmed by immunohistochemistry (IHC), the hub-genes protein levels on slides were further utilized to establishing the prognosis prediction models by several machine learning methods, including least absolute shrinkage and selection operator (LASSO), generalized linear model (GLM) and linear discriminant analysis (LDA). RESULTS There were 404 DE-IRGs determined in glioma in comparison with healthy tissue samples in TCGA. After evaluated by univariate and multivariate cox-regression, 55 DE-IRGs remaining significant. Thirty-four DE-IRGs were verified to be survival associated in both CGGA and our sequencing cohort, among which, 4 hub-genes (LIF and MDK, LAMB1, and CALU) were determined by enrichment analysis. All of the 4 hub-genes were expression confirmed on the protein level by IHC staining. Based on the staining results, a survival risk score (SRS) was established by machine learning methods (GLM, LDA, and LASSO). Based on SRS, glioma patients were divided into high- and low-risk groups, the overall survivals were significantly different between those two groups, regardless of the tumor grades or pathology types. CONCLUSIONS This study identified a potential immune prognostic signature (LIF and MDK, LAMB1, and CALU) of glioma that would be helpful for OS predicting and for potential therapeutics development.

EPCO-24. MACHINE LEARNING ALGORITHMS UTILIZING TRANSCRIPTIONAL SIGNATURES PREDICT SURVIVAL OUTCOMES IN GLIOMA