Inside the present research, i seemed and you will downloaded mRNA expression processor chip study regarding HCC architecture on the GEO database with the words from “hepatocellular carcinoma” and you can “Homo sapiens”. Half dozen microarray datasets (GSE121248, GSE84402, GSE65372, GSE51401, GSE45267 and GSE14520 (in line with the GPL571 platform) was basically acquired to own DEGs studies. Specifics of the brand new GEO datasets used in this research are provided for the Desk 1. RNA-sequencing studies regarding 371 HCC tissues and fifty normal architecture normalized because of the log2 conversion process have been gotten in the Cancer Genome sites de rencontre pour joueurs aux usa Atlas (TCGA) to own viewing brand new integrated DEGs regarding the six GEO datasets and you will building gene prognostic habits. GSE14520 datasets (based on the GPL3921 program) provided 216 HCC buildings having done scientific suggestions and you will mRNA phrase investigation to possess outside recognition of one’s prognostic gene signature. Shortly after excluding TCGA times having incomplete scientific advice, 233 HCC people the help of its complete years, intercourse, sex, tumor levels, American Combined Committee to the Malignant tumors (AJCC) pathologic tumefaction stage, vascular attack, Operating-system position and you can day suggestions was indeed provided for univariable and you may multivariable Cox regression analysis. Mutation studies was in fact taken from brand new cBioPortal to possess Disease Genomics .
Running of gene phrase analysis
To integrated gene expression chip data downloaded from the GEO datasets, we firstly conducted background correction, quartile normalization for the raw data followed by log2 transformation to obtain normally distributed expression values. The DEGs between HCC tissues and non-tumor tissues were identified using the “Limma” package in R . The thresholds of absolute value of the log2 fold change (logFC) > 1 and adjusted P value < 0.05 were adopted. Mean expression values were applied for genes with multiprobes. Then, we used the robust rank aggregation (RRA) method to finally identify overlapping DEGs (P < 0.05) from the six GEO datasets.
Framework from a potential prognostic signature
To identify the prognostic genes, we firstly sifted 341 patients from the TCGA Liver Hepatocellular Carcinoma (TCGA-LIHC) cohort with follow-up times of more than 30 days. Then, univariable Cox regression survival analysis was performed based on the overlapping DEGs. A value of P < 0.01 in the univariable Cox regression analysis was considered statistically significant. Subsequently, the prognostic gene signature was constructed by Lasso?penalized Cox regression analysis , and the optimal values of the penalty parameter alpha were determined through 10-times cross-validations by using R package “glmnet” . Based on the optimal alpha value, a twelve-gene prognostic signature with corresponding coefficients was selected, and a risk score was calculated for each TCGA-LIHC patient. Next, the HCC patients were divided into two or three groups based on the optimal cutoff of the risk score determined by “survminer” package in R and X-Tile software. To assess the performance of the twelve-gene prognostic signature, the Kaplan–Meier estimator curves and the C-index comparing the predicted and observed OS were calculated using package “survival” in R. Time-dependent receiver operating characteristic (ROC) curve analysis was also conducted by using the R packages “pROC” and “survivalROC” . Then, the GSE14520 datasets with complete clinical information was used to validate the prognostic performance of twelve-gene signature. The GSE14520 external validation datasets was based on the GPL3921 platform of the Affymetrix HT Human Genome U133A Array Plate Set (HT_HG-U133A, Affymetrix, Santa Clara, CA, United States).
The risk score and other clinical variants, including age, body mass index (BMI), sex, tumor grade, the AJCC pathologic tumor stage, vascular invasion, residual tumor status and AFP value, were analyzed by univariable Cox regression analysis. Next, we conducted a multivariable Cox regression model that combined the risk score and the above clinical indicators (P value < 0.2) to assess the predictive performance. The univariable and multivariable Cox regression analysis were performed with TCGA-LIHC patients (n = 234) that had complete clinical information.