A refined therapeutic plan based on the machine-learning prognostic model of liver hepatocellular carcinoma

Abstract

To deeply explore new strategy of the individual therapy for the patients with liver hepatocellular carcinoma (LIHC), we observed gene expression profile in patients with LIHC and made a comprehensive analysis of the inflammation-related phenotypes, we detected a set of characteristic genes associated with the biological activities of tumor cells, among which 3 genes and 2 lncRNAs are tagged on the LIHC prognosis. Then we constructed a novel prognostic model by machine learning, called Inf-PR model, and evaluated the drug sensitivity and immune targets by a series of bioinformatics tools. Ten-fold cross-validation testified that the model achieved excellent performance on prediction and classification of prognostic risks, which was not only able to get more reliable prognosis information than the age, gender, grade and stage, but also exceeded those previously reported similar models. Accordingly, drug sensitivity was detected in different prognostic risk groups, the result displayed that 10 FDA-approved small molecular drugs including lovastatin and sorafenib had higher sensitivities and perturbativities in the high-risk group, and other 15 drugs including doxorubicin and lenvatinib had better sensitivities and perturbativities in the low-risk group. Moreover, it suggested the patients with high risk would better combine with immunotherapy than those with low risk. Taken together, this study presents a new individual precision strategy about drug and target selection to treat LIHC based on this evaluation model, which is a powerful supplement for current anti-tumor therapy.