the limma user guide suggests filtering out probe sets that are not or barely expressed across conditions. In past analyses, I have simply handed all "main" probe sets over to limma, but I think it might be interesting to play around with different filtering approaches to increase power. I have found some very good posts discussing this question (e.g. https://support.bioconductor.org/p/57753/), but I would be interested to hear about people's filter habits & opinions: do you usually achieve best results with aggressive filtering, or is it better to be gentle? I know this may very much depend on the data in question... but I am still curious about the experience of others.
Many thanks in advance.