Constructing a new prognostic signature of gastric cancer based on multiple data sets

Abstract

ABSTRACT In order to explore new prediction methods and key genes for gastric cancer. Firstly, we downloaded the 6 original sequencing data of gastric cancer on the Illumina HumanHT-12 platform from Array Expression and Gene Expression Omnibus, and used bioinformatics methods to identify 109 up-regulated genes and 271 down-regulated genes. Further, we performed univariate Cox regression analysis of prognostic-related genes, then used Lasso regression to remove collinearity, and finally used multivariate Cox regression to analyze independent prognostic genes (MT1M, AKR1C2, HEYL, KLK11, EEF1A2, MMP7, THBS1, KRT17, RPESP, CMTM4, UGT2B17, CGNL1, TNFRSF17, REG1A). Based on these, we constructed a prognostic risk proportion signature, and found that patients with high-risk gastric cancer have a high degree of malignancy. Subsequently, we used the GSE15459 data set to verify the signature. By calculating the area under the recipient operator characteristic curve of 5-year survival rate, the test set and verification set are 0.739 and 0.681, respectively, suggesting that the prognostic signature has a moderate prognostic ability. The nomogram is used to visualize the prognostic sig-nature, and the calibration curve verification showed that the prediction accuracy is higher. Finally, we verified the expression and prognosis of the hub gene, and suggested that HEYL, MMP7, THBS1, and KRT17 may be potential prognostic biomarkers.