Diagnostic Models for Primary Hepatocellular Carcinoma Based on Multiple Machine Learning Methods

Liver cancer is one of the most common malignant tumors. Early diagnosis of liver cancer is difficult. The treatment risk of middle and advanced liver cancer is very high, and the prognosis is poor. Therefore, this study aimed to identify potential biomarkers closely associated with liver cancer and construct a liver cancer diagnosis model based on the machine learning. Firstly, gene expression profile data of patients with hepatocellular carcinoma and normal tissues were downloaded from the cancer genome atlas and gene expression omnibus databases. Secondly, weighted gene co-expression network analysis and differential expression analysis were used to identify shared genes. Then, gene ontology and pathway enrichment analysis were performed for shared genes. Thirdly, two machine learning algorithms, SVMRFE and XGBOOST were used to identify potential biomarkers in the shared genes. Finally, the BP, LSSVM and RF diagnosis models of hepatocellular carcinoma based on the whale algorithm (WOA) were established. The results found that 14 potential biomarkers in the shared genes were identified. By comparing three diagnostic models, the WOA-LSSVM demonstrated robust performance well on the training set, test set and validation set. The area under the curve on the training set, test set and validation set were 98%, 97 % and 96%, respectively. The WOA-LSSVM had a better diagnostic effect on hepatocellular carcinoma than other models. It provided a reference for targeted drug therapy and disease diagnosis.