Do we need to filter out control probes and low expressed probes (as given in LIMMA user guide 2016 page number 111) for all Agilent microarray platform

Question

[anandprem1792]

I am analyzing a GEO dataset of platform (GPL17077) Agilent-039494 SurePrint G3 Human GE v2 8x60K Microarray. I followed the steps given in LIMMA user guide 2016 (page number 110-112). The protocol given there is for Agilent 4x44K Arrays. So my question is whether i have to filter out control probes and low expressed probes for the "platform SurePrint G3 Human GE v2 8x60K Microarray" ?

I opened the GPL file GPL17077 and found that there are only 1 negative control and more number of postive controls (more than 50 approx).

I performed the DEG analysis in both the ways by filtering out the low expressed probes and without filtering out the low expressed probes. And I found that more DEGs are obatined when i filtered the low expressed probes.

Also, I have 10 control array and 9 test arrays. So can i keep probes at atleast 10% brighter than negative controls on 9 arrays only ?

Answer 1

[ATpoint]

Take a look at this workflow for some inspiration. In this they filter for probes with low intensity signal and probes that attract non-unique sequences (in NGS terms one would call them multimappers). Filtering out the controls should reduce the multiple testing burden from what I understand, which is probably the reason you get more significant results when using topTable. My current approach is to filter out all non-relevant probes to maximize power to call significantly different genes. For this I remove all smallRNAs, all control probes, all those with no gene assigned and those with poor intensities (the latter as described in the workflow). Still I am rather unexperienced with microarrays so others might have different approaches. Coincidently we had a discussion about exactly this topic this week on the Biostars slack (biostar.slack.com: Chat for the biostars community) which you are welcome to join and read.