Use of detection scores in bead array

Question

[dharl]

Would it make sense to remove the probes which have a low p value and bad annotation quality before doing the normalisation (quantile or neqc) across arrays?

Answer 1

[Sean Davis]

Normalization is done to make the data similar across arrays. Given that all arrays will have unexpressed genes, I doubt that doing so will significantly "improve" results, but I don't have any evidence to support my claim. Filtering is usually applied after normalization (can't think of any large-scale counter-examples) and is probably best based on variance rather than expression levels.

http://www.pnas.org/content/107/21/9546.long

Answer 2

[Gordon Smyth]

No, there is no purpose in doing so, and it might make it more difficult for the background correction and normalization algorithms to estimate the background intensity levels. As Sean says, it is much better to whatever filtering you want to do after background correction and normalization steps are completed.