Figure 3

Transcriptional diversity of 50 KEGG biological pathways within breast cancer subtypes from the Affymetrix dataset. A) Heatmap of mean dispersion distance within breast cancer subtypes considering genes in each of the 50 KEGG pathways. Pathways (rows) are ranked from the least diverse at the top to the most diverse at the bottom. Blue represents low and red high mean dispersion. B) Detailed expression heat maps for basal-like cancers showing the heterogeneity of gene expression for genes in the least heterogeneous (ribosome metabolism; top) and the most heterogeneous pathways (linoleic acid metabolism; bottom). Blue represents low and red high expression level. C) Distribution of pathway-based transcriptional diversity within each subtype. Pathway-level mean dispersion distances were calculated by bootstrap as described in Supplementary Methods. D) Comparison of pathway-level transcriptional diversity between two clinically distinct phenotypes of basal-like cancers, an extremely chemosensitive (pCR) and a chemoresistant (RD). Points on the plot represent the average pathway-level mean pairwise dispersion obtained from boostrap within each of the 50 pathways. The dashed red line is the diagonal, indicating equal transcriptional diversity between the two response phenotypes. The regression line (blue solid line) its 95% pointwise confidence interval (grey area) is consistently below the diagonal suggesting greater transcriptional diversity for RD cancers throughout the 50 pathways. Pathways that were extreme outliers from the trend described by the regression line were identified by quantile-quantile plots of the standardized residuals. These pathways are indicated with letters as following: A – sphingolipid meta, B – SNARE interactions in vesicular transport, C – basal cell carcinoma, D – dorso-ventral axis formation, E – non-homologous end joining (DNA repair), F – folate biosynthesis.