Abstract: Objective To construct the prognostic evaluation model for gastric cancer (GC) and identify biomarkers related to the initiation and prognosis of GC through bioinformatics analysis of TCGA dataset. Methods The transcriptome data of GC and corresponding clinical information (containing 375 GC samples and 32 para-carcinoma samples) were downloaded from the TCGA website and merged into a matrix. The differential expressed genes (DEGs) were screened between GC and adjacent normal tissues using R“edgeR”package, and Cox univariate and multivariate regression analyses were performed by R“survival”package, and the predictive signature model of GC prognosis was established. Combined with clinicopathologic parameters, the role of the gene signature in predicting prognosis of GC was validated. Results A total of 4 332 genes were regarded as DEGs based on“edgeR”package. Next, Cox univariate regression analysis screened 710 DEGs associated with prognosis, among which 25 DEGs were incorporated into Cox multivariate analysis on the threshold of P value＜0.001. Consequently, we obtained an 8 DEG (BCHE, INPP5J, VCAN, IGFBP1, CGB5, HP, PSG9,AFF2) predictive model for GC prognosis. Based on the median value of risk score, GC samples were divided into high and low risk groups. Kaplan-Meier survival curve indicated that the 5-year survival rates in high and low risk groups were 56.20% and 17.27%, respectively (P＜0.001). ROC curve confirmed the medium accuracy of the predictive model in GC prognosis. Furthermore, the 8 gene signature was proved to be an independent prognostic factor in GC when clinicalpathologic information was incorporated into the Cox multivariate regression model. Conclusion Prognostic evaluation model of gastric cancer based on bioinformatics method can become new indicators for prognosis of gastric cancer.

Key words: stomach neoplasms, prognosis, gene expression profiling, computational biology