Identification of a three-miroRNA-based prognostic gene signature in hepatocellular carcinoma using the random survival forest method

Background: Hepatocellular carcinoma (HCC) is the fourth leading cause of cancer-related deathworldwide. HCC has poor prognosis and monitoring of HCC patients with high risk is necessary. Manystudies have revealed miroRNAs (miRNAs) as potential prognostic biomarkers in cancer. For this aim, weperformed a bioinformatic-based analysis and random survival forest method to identify a prognostic genesignature in HCC.Materials and Methods: miRNA-seq data and clinical information of HCC samples were downloaded fromThe Cancer Genome Atlas (TCGA) database using the TCGAbiolinks R package. Differentially-expressedmiRNAs (DEMis) among cancerous and paracancerous samples were identified by the DESeq۲ packagebased on |logFC| > ۲ and adjusted p-value < ۰.۰۱. Univariate survival analysis was performed on the DEMisdata of cancerous samples to identify prognosis-related miRNAs (hazard ratio (HR) ≠ ۱ and p-value < ۰.۰۱)by the survival package. Subsequently, randomForestSRC package was utilized to rank survival-relatedmiRNAs. Then, multivariate cox regression analysis was performed to establish a risk scoring model basedon the regression coefficient and gene expression. Survival and time dependent ROC (receiver operatingcharacteristic) curve plots were generated by the survival and survivalROC packages, respectively.Results: ۱۳۱ DEMis (۱۲۶ upregulated and ۵ downregulated) were identified between ۳۷۲ cancerous and ۵۰paracancerous HCC samples. Univariate survival analysis revealed ۸ prognostic-associated miRNAs. Basedon the random forest analysis, ۳ miRNAs (hsa-miR-۹-۵p, hsa-miR-۱۳۷-۳p and hsa-miR-۱۰۵-۵p) which hadthe relative importance > ۰.۶, were selected to construct a prognostic gene signature. The HR and p-value ofthe model were ۲.۶ and ۰.۰۰۰۰۳, respectively. Additionally, AUC of the model for ۵, ۳, and ۱ year were ۰.۶۶,۰.۶۱, and ۰.۶۴, respectively.Conclusion: This study provides a reliable gene signature for prediction ofprognosis in HCC patients using miRNAs. Further studies are needed to evaluate the strength of this modelin HCC.