Abstract
The present study aimed to establish a prediction model for hepatocellular carcinoma (HCC) based on the cross talk genes from important biological pathways involved in HCC. Differentially expressed genes (DEGs) for HCC were identified from mRNA profiles of GSE36376, which were mapped to protein‑protein interaction (PPI) networks from BioGrid and the human protein reference database. Then critical genes based on the deviation score and the degree of node were selected from the novel PPI network. Cross talk genes were screened from the network established based on the associations of gene‑gene, gene‑pathway and pathway‑pathway. A classifier based on specific cross talk genes was constructed for prediction of HCC using the random forest algorithm. Finally, the diagnostic performance of this prediction model was verified by predicting survival time of patients with HCC from the genome cancer atlas (TCGA) and other independent gene expression omnibus (GEO) databases. From the novel PPI network, a total of 200 critical genes were screened out and they were significantly enriched in 23 pathways, which have been reported to be significantly associated with the development of HCC. Based on these identified pathways, cross talk genes were identified including AKT1, SOS1, EGF, MYC, IGF1, ERBB2, CDKN1B, SHC2, VEGFA and INS. The prediction model has a relative average classification accuracy of 0.94 for HCC, which has a stable predicting efficacy for survival time of HCC patients validated in the TCGA database and two other independent GEO datasets. In conclusion, a total of 39 cross talk genes in HCC were identified and a classifier based on the cross talk genes was constructed, which indicates a high prognosis prediction efficacy in several independent datasets. The results provide a novel perspective to develop a multiple gene diagnostic tool for HCC prognosis, which also provided potential biomarkers or therapeutic targets for HCC.