Fig. 3.

Subtype identification from batch corrected data. Hierarchical clustering of batch-corrected genomic data from probes with maximum intratumor heterogeneity. Samples are colored according to the processing technique by which they were collected (blue for frozen samples processed with Ovation amplification kits, red for FFPE samples processed with FFPE amplification kits and green for frozen samples processed with FFPE amplification kits) and labeled according to the subtype (i.e. basal, mesenchymal, atypical and classical) inferred by applying the Walter et al. (2013) classifier to the batch-corrected data. Clustering is performed on data (a) without batch correction, (b) with ComBat, (c) with SVA, (d) with SVA combined with ComBat, and (e) with pSVA