Fig. 1

EMAT clusters and their characteristics. a EMAT clusters based on lymph node-negative METABRIC samples obtained using hierarchical clustering. The heatmap shows the normalized expression of EMAT genes (rows) in each sample (columns). Sample dendrogram colors are chosen to match those of Kaplan-Meier plots in c. b Characterization of samples based on similarity to hESC, PAM50 subtypes, ER, PR, and HER2 status, stage, grade, and type of treatment. Spearman’s rank correlation, scaled between 0 and 1 using min-max normalization, is used as the measure of similarity of samples to hESC, in which 0 and 1 represent least similar and most similar, respectively. c Kaplan-Meier plots corresponding to n = 4 clusters. The heatmap shows the relative ranking of the average expression of four biomarkers in each cluster compared to other clusters. Clusters EMAT2 and EMAT3 have very similar Kaplan-Meier survival curves (c), even though their gene expression profiles are distinct. EMAT4 has a worse survival outcome compared to other clusters (p = 0.05 against EMAT3, p = 0.01 against EMAT2 and p = 1.8E-6 against EMAT1). d The box plots show the distribution of hESC similarity of the samples in each cluster. The similarity is defined as the Spearman’s rank correlation (scaled between 0 and 1) between expression profiles of H1 hESC lines and each sample. The p-values (calculated using a one-sided t-test) show how significant the differences between two adjacent EMAT clusters are with respect to their similarity to hESC. The significance p-value for the cluster with the least similarity to hESC (EMAT1) and the cluster with the most similarity to hESC (EMAT4) is p = 1.7E−23