Figure 2. Characterized chromatin-remodeling factors selection based on random forest algorithm. (A) The relationship between the number of variables and model accuracy after multiple cross-validations of the random forest process was shown. The auxiliary line was added at 0.9 accuracy and the first point indicated 2 variables while the following points added one variable compared to the previous one. (B) The importance of each variable in model contribution was shown. The horizontal axis represents the position of 86 chromatin-remodeling factors in the rank of high to low importance from right to left. Red features were confirmed unimportant, yellow features were designated as tentative and green features were confirmed important. Three blue boxplots mapped to minimal, mean and maximum Z scores of shadow attributes respectively. (C) Z score gene expression level of 86 chromatin-remodeling factors among consensus cluster A and cluster B. Clinicopathological stage, outcome status and groups (cluster C and cluster D) of consensus cluster based on BPTF, CNOT1, and SIN3A were annotated in the heatmap. (D) Comparison of overall prognostic differences between Cluster C and Cluster D upon all stage patients. (E) Comparison of progression-free prognostic differences between Cluster C and Cluster D upon all stage patients.