Construction of DNA methylation-based biomarkers for diagnosis of hepatocellular carcinoma

3048 Background: Hepatocellular carcinoma is the most common malignant tumor in China. Most of the patients are found in the advanced stage, which causes its low survival rate. According to the latest epidemiological data, hepatocellular carcinoma is the second leading cause of cancer mortality. DNA methylation is closely related to tumor development. In this study, differentially expressed methylated genes related to tumorigenesis and development were screened from the database and the diagnostic model was constructed for early screening. Methods: Limma package was used to analyze the differences between TCGA _ LIHC liver cancer and GSE113409 normal blood methylation sites and differentially expressed methylation sites (logFC>1，p<0.05) were identified. The TCGA _ LIHC methylation database was used as the training set. Lasso method, and random forest method were used for screening the differentially expressed methylation sites for the construction of the diagnostic model. After that, the characteristic coefficient of methylation sites and ROC curve of the diagnostic model were obtained from the training set. According to the cd-score, HCC patients could be diagnosed in the early stage, thus improved their survival. Results: Limma package was used to analyze the differences between TCGA _ LIHC liver cancer and GSE113409 normal blood methylation sites and 7109 differentially expressed methylation sites (logFC>1，p<0.05) were identified. Then the top 1000 differentially expressed methylation sites were selected. The TCGA _ LIHC methylation database was used as the training set. 30 methylation sites were screened by lasso method, and 20 methylation sites were screened by random forest. Taking the intersection, a total of 6 methylation sites were obtained for diagnostic model: cg00346208, cg00084798, cg00458878, cg00006397, cg00325910, cg00399027. The characteristic coefficient of methylation sites were displayed in the table below and the AUC of the diagnostic model was 0.9991579. High cd-score tended to have hepatocellular carcinoma, and vice versa. Conclusions: We have identified several DNA methylation markers for the diagnosis of hepatocellular carcinoma from GEO and TCGA databases. This model can predict hepatocellular carcinoma and the AUC is higher than that of AFP, which means it may become a more sensitive indicator for the diagnosis of liver cancer. [Table: see text]