Fig. 8.

Deep feature patterns within the six breast cancer subtypes. The columns are the 842 patients ordered by the deep feature-based six subtypes and the rows are the 500 deep features. The feature clusters were assigned using the hierarchical clustering with complete linkage and Euclidean distance. The top horizontal annotation bars show (from top to bottom): deep feature-based BC subtype, DCNN_Concat predicted subtype, PAM50 mRNA expression subtype, ER/PR/HER2 expression status-defined subtype, ER IHC expression status, HER2 copy number state, age (>= 45 as defined as old, <45 as young) and histological grade. Details on color coding of the annotation bars are presented below the heat map.