Research Paper Volume 16, Issue 14 pp 11248—11274

Figure 7. Establishment and validation of the stemness subtype-based risk signature in TCGA and cohort in our hospital. (A) The performances of three machine-learning algorithms (LASSO, COX and RF) for feature selection were, respectively, evaluated in the training set and validation set. AUCs were generated by ROC analysis. (B) Venn diagram showing the common genes of the three machine-learning methods. (C) Left panel: confusion matrices of binary results of the Stemness Subtype Predictor for the training set (upper) and validation set (lower). Right panel: ROC curves of the Stemness Subtype Predictor in distinguishing two subtypes in the training set (Upper, AUC = 0.960) and validation set (Lower, AUC = 0.922). (D, E) The heatmap showing the expression levels of 7 hub genes in the subtype I and subtype II. The distribution of clinicopathological features was compared between the two groups in TCGA cohort and cohorts in our hospital, respectively. (F) Kaplan-Meier curve of patients in low- and high-risk groups of OS and DFS in TCGA patients. (G) Kaplan-Meier curve of patients in low- and high-risk groups of OS and DFS in patients in our hospital.