Dear list,


I am working with Affymetrix Human Gene 1.1-ST Arrays. My data is pre-process with RMA in the oligo Package and the limma Package is utilized to assess differential expression (lmFit and eBayes). 


I get a high number of DE genes. Most of these belong to Category: Main. However, some of them, belongs to other categories such as normgene->intron, control->affx, normgene->exon. I have the same problem also when pre-processing with Plier (we get data from core facility pre-processed with Plier) and/or doing the statistics in SAM. Of course, it's easy to remove everything that do not belong to Main in the DE list.


As I understand it, empirical Bayes takes into consideration the full 
data set. Removing all other categories, some of them which are not 
supposed to be changed between experiments, before statistics will 
affect the outcome of the test (DE list). However, I have never been 
interested in the other categories to start with. 


So, my question is what your stand point is regarding filtering out everything that do not belong to Category: Main before doing statistics. 



Best regards
Martin O'Hare
	[[alternative HTML version deleted]]