Fig. 6

Genomic selectivity of ERα/FOXA1 action classifies male breast cancers on outcome. a Unsupervised hierarchical clustering of ERα and FOXA1 peak intensities at sites that classify on negative (yellow) or positive (orange) LN-status. b Unsupervised hierarchical clustering of the potential target genes driven from ERα and FOXA1 binding sites that classify on LN-status (discovery cohort). Negative (yellow) and positive (orange) LN-status, M1 (red), and M2 (blue) subtypes are indicated by the color bars above. c The same analysis as in b, using the validation cohort. Negative (yellow) and positive (orange) LN-status, M1 (red) and M2 (blue) subtypes as well as development of distant metastasis (red and green) are indicated by the color bars above. d Receiver-operator characteristic (ROC) curve indicating predictive performance of binary classification models trained with potential target genes of ERα (blue), FOXA1 (green) or the union of the two (purple). e Area under curve of the ROC curves of d, with p-values from bootstrapping analysis indicated on the top of each bar. f Kaplan-Meier analysis of two groups of patients (high- and low-risk groups in red and black lines, respectively) driven from the classification model trained with union of potential targets. Difference in survival is assessed with log rank test (p-value = 0.048). g Bar plot indicating 14 genes contributing to the classification model and their coefficients in the model. Predicted upstream regulator (blue − ERα and green − FOXA1) and sign of the coefficients (red − negative, green − positive) are indicated