Clinical samples processed on Affymetrix and Illumina platforms.A) Experiment Layout of the overlapping 18 matched clinical breast cancer samples from 6 patients from similar Affymetrix and Illumina datasets studying the of the effect of Letrozole in the neoadjuvant setting. B) Pairwise Pearson correlation heatmaps (left) from 18 matched clinical breast cancer samples from 6 patients demonstrate cross platform bias and the effects of three correction methods, mean-centering, distance-weighted discrimination (DWD), an Empirical Bayes method (ComBat) and cross platform normalisation (XPN). R values range from low correlation (red) to high correlation (white) through shades of orange and yellow reflecting the overall similarity of expression profiles based upon biological and platform-specific variation. The inner diamond represents the matched samples from the two platforms. Each patient sample is numbered as untreated (−1), 14 days (−2) and 3 months (−3) post treatment. Uncorrected data is NOT shown (to show it on the same colour scale as the other plots would not demonstrate the differences between the correction methods).
Turnbull et al. BMC Medical Genomics 2012 5:35 doi:10.1186/1755-8794-5-35