Random-forest algorithm based biomarkers in predicting prognosis in the patients with hepatocellular carcinoma

Hepatocellular carcinoma (HCC) one of the most common digestive system tumors, threatens the tens of thousands of people with high morbidity and mortality world widely. The purpose of our study was to investigate the related genes of HCC and discover their potential abilities to predict the prognosis of the patients.

We obtained RNA sequencing data of HCC from The Cancer Genome Atlas (TCGA) database and performed analysis on protein coding genes. Differentially expressed genes (DEGs) were selected. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment were conducted to discover biological functions of DEGs. Protein and protein interaction (PPI) was performed to investigate hub genes. In addition, a method of supervised machine learning, recursive feature elimination (RFE) based on random forest (RF) classifier, was used to screen for significant biomarkers. And the basic experiment was conducted by lab, we constructe a clinical patients' database, and obtained the data and results of immunohistochemistry.

We identified five biomarkers with significantly high expression to predict survival risk of the HCC patients. These prognostic biomarkers included SPC25, NUF2, MCM2, BLM and AURKA. We also defined a risk score model with these biomarkers to identify the patients who is in high risk. In our single-center experiment, 95 pairs of clinical samples were used to explore the expression levels of NUF2 and BLM in HCC. Immunohistochemical staining results showed that NUF2 and BLM were significantly up-regulated in immunohistochemical staining. High expression levels of NUF2 and BLM indicated poor prognosis.

Our investigation provided novel prognostic biomarkers and model in HCC and aimed to improve the understanding of HCC. In the results obtained, we also conducted a part of experiments to verify the theory described earlier, The experimental results did verify our theory.