Construction of a prediction model for hepatocellular carcinoma based on machine learning and its prognostic...

Hepatocellular carcinoma (HCC), as a cancer with high morbidity and mortality, urgently requires the development of a clinical prediction model with high robustness and generalizability and its prognostic study of the tumor microenvironment to provide personalized clinical treatment for patients. Key prognostic genes were screened by analyzing mRNA expression data from GTEx and The Cancer Genome Atlas (TCGA) using limma difference analysis, Cox analysis, and machine learning (ML) algorithms. TCGA database was used as a training set, and the International Cancer Genome Consortium database was used as a test set to screen the best prognostic modeling algorithms using a combination of 101 ML algorithms for training and constructing Nomo score plots based on the algorithmic risk scores as well as Shiny online prediction models. Based on shapley additive explanations analysis, drug sensitivity analysis, and immune infiltration analysis were performed on the 6 genes screened to visualize the importance of prognostic genes. HCC tumor mutation load analysis was also performed. A risk prediction model for HCC death was developed based on the RSF algorithm, with an RSF model C-index of 0.765 and AUC values of 0.978, 0.989, and 0.964 for 1-, 3-, and 5-year ROC curves for the Nomo score model, respectively. LPL, RAET1E, RNASEH2A, GTF2H4, SCML2, and PRDM12 were potential diagnostic and prognostic markers, among which SCML2 and PRDM12 were significantly correlated with multiple drugs in drug sensitivity analysis.TP53 mutations were correlated with patients' age, chronological age, gender, histological tumor stage, T stage, and lymph node metastasis. An online HCC mortality risk prediction model was developed using the RSF algorithm. LPL, RAET1E, RNASEH2A, GTF2H4, SCML2, and PRDM12 are potential prognostic target genes, whereas TP53 mutations are associated with clinical features that may inform the development of HCC therapy.