[BioC] Loess Normalization vs. Dye-Swap Normalization

[Ulisses M. Braga-Neto]

Hello:

Assume that one conservatively expects to see a dye effect in Agilent
data for all genes. One should then for example average expression values
of corresponding green and red channels in a dye-swapped design.

The routine "loess.normalize" in the Affy package does normalization
between all pairs of arrays. Assuming there is a dye effect, would it make
any sense to use individual red and green channels in this process? Or
should one first do dye normalization and then apply pairwise loess
normalization?

Another intriguing question (at least for me) is whether one should take
logs prior to loess normalization. I know why one should usually take logs
to do t-test or ANOVA analysis (to make the data look more Gaussian), but
it is not clear if this should be done before or after applying loess.

Thank you for your time.
Ulisses.

------------------------------------------------------------------------
Ulisses M. Braga-Neto, Ph.D.       Dept. of Electrical and Computer Eng.
Assistant Professor                                 Texas A&M University
ulisses at ece.tamu.edu                                 Tel: (979) 862-6441
http://www.ece.tamu.edu/~ulisses                     FAX: (979) 845-6259