Fig. 2

PAM50 and feature gene classifications group canine and human basal-like tumors together, but separates canine non-basal-like tumors from human tumors. A. PAM50 classification of 143 subtyped canine mammary tumors. The heatmap shows hierarchical clustering of the canine tumors using the row-scaled log₂(TPM) values of 43 out of the 50 PAM50 genes. The bars indicate the canine subtype from Fig. 1, as well as the PAM50 subtype and confidence score of each tumor. B. An example of cross-species PAM50 classification. All 143 subtyped canine tumors, together with 267 human tumors (60 tumors per subtype of luminal A (LumA), luminal B (LumB), basal-like, or HER2-enriched randomly sampled from TCGA database, along with all 27 normal-like tumors from TCGA), were subjected to PAM50 classification. The dendrogram is colored to indicate the minimum number of clusters that maximally separate the hLumA tumors from hBLBC tumors. This cross-species PAM50 classification was repeated 100 times. C. Bar plot showing the number of random samplings in which human and canine basal-like tumors clustered together or separately, compared to those of human luminal A and canine non-basal-like tumors. The p-value is based on Fisher’s exact test. D. Multidimensional scaling plot of the cross-species PAM50 classification shown in B. Each dot represents a tumor from a subtype specified by the color as indicated in the legend. E. Violin plot indicating the distribution of the Mahalanobis distances between the centers of two subtypes on the multidimensional scaled plot D of each of the 100 cross-species PAM50 classifications B achieved via random samplings (see Methods). The p-values were obtained from Wilcoxon tests. F. Classification of 143 subtyped canine mammary tumors using 115 feature genes for hBLBC and 130 feature genes for hLumA identified via machine learning. G. Violin plot indicating the distribution of the Mahalanobis distances as described in E, using the 115 hBLBC feature genes or 130 hLumA feature genes for cross-species classification, instead of the PAM50 genes. The p-values were obtained from Wilcoxon tests.