Figure 4

Distributions of t-statistics calculated from differently normalized MAS5 expression values in the Breast Cancer data. The normalization procedures applied were the same as in Figure 3: global mean as default and housekeeping normalization before and after taking logarithms (raw and log respectively in the legend.) The t-statistics compare expression values between users and non-users of hormone replacement therapy (HRT) in the sample. The density curves show the smoothed histograms of the test statistics for all 22283 probe sets on the chip. The default curve is centered at zero, indicating balanced up- and down-regulation in reaction to HRT, whereas the curves for the housekeeping-normalized data are shifted to the left, indicating a massive down-regulation of thousands of genes in HRT users, which is biologically much less plausible. This effect is more pronounced for the raw than for the logged normalization, same as with the residual correlations shown in Figure 3.