Figure 3

Validation of the luminal A and basal-like subtypes using three microarray platforms. (A) Unsupervised hierarchical clustering of 20 breast tumor tissues analyzed by Applied Biosystems (two replicates per sample), Stanford cDNA and Agilent arrays using the 510 mapped intrinsic genes. The data from each platform were first transformed independently and then combined for clustering: the level of expression of each gene in each sample (Applied Biosystems microarrays: normalized signal intensity; Stanford cDNA and Agilent microarrays: normalized log2 ratio of the sample vs. the reference (UHR)) was transformed into a log2 ratio relative to the median level of expression of that gene across all the samples within the data set of the given platform. The experimental dendrogram displays the clustering of the tumors into distinct subgroups. Branches are color-coded according to the subtype with which the corresponding tumor sample showed the highest correlation. Tumors with low correlation (< 0.2) with a specific subtype are indicated by gray branches. Luminal A subtype (dark blue) and basal-like subtype (red). (B) Venn Diagram of the most differentially expressed genes identified by all three different array platforms; 319 genes were identified as the common signature genes using ANOVA analysis and the following criteria: (1) > 2-fold change between the two subtypes, and (2) False discovery rate < 5%. (C) PCA analysis of luminal A and basal-like samples using a minimum set of 54 genes identified by PAM analysis. Data sets generated from three array platforms (48 arrays total, two replicates per sample in the Applied Biosystems data set) on 6 luminal A and 6 basal-like tumor samples using the 319 common signature genes were used as training set for the PAM analysis.