I have to analyze a large number of Affymetrix microarray datasets, something I haven't done in many years. I need to select the most informative/appropriate probeset to do DE analysis on. I understand this might not be optimal, but it's needed here.

After reading up I'm confused about the _x_at probesets. Why are they included if not guaranteed to be uniquely mapping to one gene? Is it standard practice to exclude them? Would a reasonable selection process be:

1. Exclude _x_at probesets
2. Exclude probesets with > 50% of samples in all tested groups (always just case vs control) having absent calls
3. Use the probeset with highest median signal in any of the groups being compared

I'm not interested in transcripts here, only gene-level. Would it be better to just choose _at probesets instead of the last step? Or to take the one with the smallest CV within each group?

Had forgotten what a pain MA is :)

EDIT: Some more info. I only have RMA processed data and the probesets have already been mapped to gene and is something I can't change. It's purely the ranking of probesets I'm interested in. Or pooling if that is considered better. I understand this has been asked a million times, but most threads are old and I wasn't able to find something definitive.