FIG. 7.

Proposed analysis scheme. The analysis should start with checking the background variability and if that is low and no spatial artifacts are present on the microarray image, no background correction is advised. Otherwise, a background correction is applied. After using two normalization methods of choice (in our case loess and vsn) and applying a statistical model, the intersection of two differentially expressed gene lists (each coming from one of the normalizations) is used to obtain more robust differentially expressed gene lists.