Fig. 5

Molecular subtype validation in another independent cohort. A The contribution of six signatures to the subtype diagnostic model. B Receiver operating characteristic (ROC) curves with reported areas under the curve (AUCs) demonstrated the efficacy of the subtype diagnostic model in identifying subtypes of ESCC. C A abundance of the six signatures in three subtypes. Data presented as mean ± s.d.; P values by Wilcoxon rank-sum test. D IHC and intensity statistics of four characteristic proteins in the predicted S3 subtype (n = 12) and S1/2 subtype (n = 40) in the independent ESCC cohort 2 (P value from Wilcoxon rank-sum test). Data presented as mean ± s.d.; P values by Wilcoxon rank-sum test. E, F The intensity of creatine (E) and 2’DG (F) in the predicted S3 subtype (n = 12) and S1/2 (n = 40) subtype in cohort 2. Data presented as mean ± s.d.; P values by Wilcoxon rank-sum test. G Kaplan–Meier curves for overall survival analysis of predicted S3 subtype and S1/2 subtype (P value from the log-rank test). H, I Spearman correlation analysis of the protein expression of HK3 protein and the intensity of creatine in our study (H) or in cohort 2 (I). J ROC curves with reported AUCs demonstrated the efficacy of the model containing only creatine and HK3 protein. K, L Kaplan–Meier curves for overall survival analysis of the expression of HK3 protein (K) and the intensity of creatine in our study (L) (P value from log-rank test). See also Table S10