Abstract:

It is a current research hotspot to establish and validate the prognosis related model of hepatocellular carcinoma （HCC） using methylation CG sites. On the R language platform， methylomics data of TCGA-LIHC and GSE136380 were analyzed. Limma package was used to analyze the differential methylation CG sites of TCGA-LIHC and GSE136380， and wgcna package was used to analyze the weighted gene co expression network of GSE136380. The intersection of the three was determined as the key methylation CG site set， and then 83 methylation CG sites were obtained. LASSO-COX risk proportion regression analysis was used to screen six methylation CG sites， including cg15744128， cg24282479， cg17922226， cg13090969， cg23505966 and cg08708079， to construct and validate the prognosis model of HCC after surgical resection. The AUC of the fitting model in training concentration for 1， 3 and 5 years is 0.756 （95% CI：0.644~0.876）， 0.749 （95% CI：0.638~0.854） and 0.797 （95% CI：0.661~0.920） respectively. The AUC of the test set for 1， 3 and 5 years were 0.848 （95% CI： 0.680~0.990）， 0.833 （95% CI：0.697~0.974） and 0.685 （95% CI： 0.553~0.903）， respectively. Kaplan Meier curve showed that it could accurately identify high-risk patients with poor prognosis （logrank test P<0.001） and patients with easy recurrence （logrank test P<0.001） after HCC resection， which is helpful to further develop the treatment plan and benefit the patient.

Key words: hepatocellular carcinoma, public database, prognostic evaluation, methylation, nomogram