Fig. 4

Decision tree and random forest analysis for different pathologic stages of HCC. a We combined the clinical, mutation and expression information to perform the random forest modeling analysis. Multiple dimension scale plot was provided. b ROC curve was plotted, and AUC value was calculated. c Decision tree modeling analysis was performed. d-f We excluded the TNM information to complete the random forest and decision tree modeling, again. g-h Based on the principles of mean decrease accuracy and mean decrease Gini, we identified the largely contributed genes. i Expression status of these genes in 50 HCC tissues with adjacent non-tumor tissues. j Expression data of stage I, II, III, and IV through the GEPIA2