Fig. 2

Enrichment of EMAT clusters in other breast cancer subtypes and systematic comparison of their prognostic power using cross-validation. The heatmaps show the − log10 (p-value) of enrichment of EMAT clusters in different subtypes or clinical parameters (using a hypergeometric test). The scatter plots compare the performance (measured in C-index) of Cox regression predictions using EMAT cluster status and clinical parameters (y-axis) versus other predictors (x-axis) (see the “Methods” section for details). If there are more points above the diagonal line (and further away from it), it shows that the method represented on the y-axis outperforms the method represented on the x-axis. The p-values were calculated using a one-sided Wilcoxon signed rank test and represent the significance of the improvement obtained using EMAT cluster status and clinical parameters as compared to other predictors. The measure PIF shows the percent of times in which EMAT + clinical parameters provided a more accurate prediction compared to the baseline. a The heatmap shows enrichment of EMAT clusters by samples of different tumor sizes. The scatter plot shows performance of Cox regression predictions using EMAT + clinical parameters versus clinical parameters alone. b The heatmap shows enrichment of EMAT clusters by samples of different PAM50 molecular subtypes. The scatter plot shows performance of Cox regression predictions using EMAT + clinical parameters versus PAM50 subtypes + clinical parameters. c The heatmap shows enrichment of EMAT clusters by samples of different receptor status. The scatter plot shows performance of Cox regression predictions using EMAT + clinical parameters versus receptor status + clinical parameters. d The heatmaps show the distribution of patients in each PAM50 subtypes as well as different treatments in EMAT clusters