[BioC] arrayQualityMetrics: expression or probe-level data?

[Efthimios MOTAKIS]

Hello all,

I have a question regarding "arrayQualityMetrics" and I could not find  
relevant information at the posted Q/A.

I have a dataset of 80 U133A Affymetrix .cel files coming from an  
ovarian cancer study (patients' samples). When I run  
"arrayQualityMetrics" to the background corrected (RMA) and normalized  
(quantiles) probe-level data (thus processing the non-summarized  
AffyBatch object) I get a PCA showing two distinct groups (not a  
random separation; the first 40 cells belong to one group and the  
other 40 to the other). I also get 2 outliers (based on 5 out of the 6  
different plots provided).

When I run "arrayQualityMetrics" to the RMA or gcRMA expression data  
the two groups effects on PCA and the outliers disappear completely.

Can summarization have such a strong influence to my data? Should I  
rely on the .cel files or the expression results for my analysis

Thank you,
Makis