Hi,

I'd like to apply a Variance Based Filtering (using the varFilter function from genefilter R package) to an ExpressionSet object whose (array quality) should be weighted with arrayWeights function. However, I'm wondering which step I must perform first, the weighting or the filtering?

Thanks

You should absolutely not be using varFilter if you are planning to use the limma package. That completely invalidates the empirical Bayes hyperparameter estimation. If you want to filter (not super critical IMO for array data, but your choice) you should use something that A) doesn't involve restricting the range of within-gene variance, and B) doesn't use any information about which group each sample comes from.

For RNA-Seq there is a valid rationale for eliminating genes with low counts, but for microarray data it's way more complicated than that due to the effects of GC content on binding. For example, if GC content gets high enough, an Affy probe will have very high binding even if the probe is not complementary to any known sequence.