Effects of rescaling, quantile normalization and a different summary procedure on subsequent analysis. Intermediate time points were interpolated using a simple linear procedure using neighbouring time points (for example, the 8 hours time point was interpolated using the 6 and 10 hours time points). A relative interpolation error was then computed from each gene using the formula (X
is the true value and X
the predicted one. In each plot there are three curves which, from top to bottom correspond to the third quartile, median and first quartile of the relative interpolation error. This procedure was repeated for each replicate separately before (A), after (B) rescaling, after quantile normalization applied on MAS5 summarized values (C) and on RMA summarized data (D). In (B) scaling imbalances have been corrected whilst the performance of quantile normalization and RMA summarized data are inferior in that respect (C, D). Only those genes whose neighbouring time points have a detection p-value less than or equal to 0.001 were included, which corresponds to 2, 500 – 5, 000 genes/time point.