Fig. 8

MMTV-Myc histological subtypes are representative of different human breast cancer intrinsic subtypes. (A) t-distributed stochastic neighbor embedding (t-SNE) performed on human METABRIC and mouse model gene expression samples using the 32 homologous gene subset of PAM50 as determined by recursive feature elimination with cross-validation. (B) Normalized scoring metrics for the soft voting classifier including accuracy for measuring true positives, a weighted F1 scoring metric for balancing precision and recall, and a Matthews correlation coefficient (MCC) metric for taking into account false positives and false negatives even in the case of unbalanced classes. (C) A confusion matrix where true positives lie along the diagonal from top left to bottom right and false values occupy all other boxes. (D) Bar chart of proportional probabilities of each model representing human intrinsic breast cancer subtypes as determined by the soft voting classifier. Human intrinsic subtype proportion was determined directly from proportions of METABRIC breast cancer patients subtyped. (E) Bar chart of MMTV-Myc histological subtypes and proportional probabilities of each subtype corresponding to different human breast cancer intrinsic subtypes determined by the soft voting classifier