Prognosis Prediction Of Hepatocellular Carcinoma Related To Immune Infiltration Based On Weighted Gene Co-expression Network And Machine Learning Algorithm

Hepatocellular carcinoma(HCC)is the second leading cause of cancer-related mortality worldwide.In recent years,the number of deaths caused by liver cancer has increased year by year.Since HCC has no specific clinical features in the initial stage,patients can usually only be diagnosed in the middle and late stages.Although surgical treatment has made great progress in liver cancer,the survival rate of the entire HCC patient population is still not satisfactory due to the high metastasis and high recurrence of patients.In addition,tumor-infiltrating immune cells,as an essential part of immunotherapy,have gradually attracted people's attention.However,there is still a lack of systematic and comprehensive research on the correlation between immune infiltration and liver cancer.Therefore,this project aims to use a variety of algorithms to develop potential immune-related markers to improve the prognosis of liver cancer patients.First,this study integrated the clinical and expression data of TCGA and GEO,and screened out 4,072 differentially expressed genes(DEGs),including 2796 up-regulated genes and 1,276 down-regulated genes.Using ESTIMATE algorithm and WGCNA algorithm,289 immune-related genes(IRGs)were defined.The KEGG pathway indicates that IRGs are involved in the signal transduction pathway of T cell receptors,hepatitis C,and the signal transduction pathway of B cell receptors.Secondly,we used three machine learning algorithms: LASSO,random forest,and diversified COX regression to construct and verify new 7-IRGs signatures(DPYSL4,FHL3,LGALS3,SLC6A3,SOX11,ST6GALNAC4,and STK32B)with excellent prognostic independence.The risk score based on the prognostic signature has an excellent ability to identify high-risk patients and predict overall survival(p <0.001;1-year AUC = 0.797;3-year AUC = 0.787;5-year AUC = 0.806).Interestingly,the nomogram formed by combining the risk scores further improved the accuracy of survival prediction(C index =0.761).In addition,gene set enrichment analysis(GSEA)analysis found that high-risk patients are significantly related to the Cytokine-cytokine receptor interaction,the Human immunodeficiency virus 1 infection,and the Human cytomegalovirus infection.In addition,the FHL3 classifier is defined using the machine learning SVM algorithm.We confirmed for the first time that high FHL3 expression corresponds to the poor prognosis of HCC patients(p = 0.00029),and with the increase of FHL3 expression,the T,grade,and stage stages show an upward trend.Univariate and multivariate COX proved to be independent prognostic factors for HCC patients.In addition,the mutation frequency in the FHL3 high expression group was significantly higher than that in the FHL3 low expression group,and the mutations in TP53,TTN and CTNNB1 were the most significant.Finally,the CIBERSORT algorithm evaluates the immune fluctuation of the HCC microenvironment.We found that T cells gamma delta,Monocytes,T cells regulatory(Tregs),Macrophages M0 and M1 have significant differences in the high and low expression groups.COX regression confirmed that the abnormal infiltration of immune infiltrating cells driven by FHL3 may be one of the important immune phenomena in the process of carcinogenesis.Interestingly,the BPNN model further confirmed this result.In summary,we integrated a variety of machine learning algorithms to construct and verify the new 7-IRGs signature risk model and FHL3 gene are expected to become potential prognostic markers of HCC.