Hi all,

I'm currently analysing a microarray dataset, on which I'm trying to detect the difference of gene expression between responders and non responders to a treatment.

Even after reading the Limma user guide and searching on the forum I'm not really sure which design matrix I should use for the comparison.

Here is a simplified model of my dataset :

cy3                 cy5
responder           RNA_reference
RNA_reference       responder
non_responder       RNA_reference
RNA_reference       non_responder
moderate_responder  RNA_reference
RNA_reference       moderate_responder

For example, I would like to detect differentialy expressed genes between responders and non responders, without taking moderate responders arrays into account. I think I understand it corresponds to a "separate channel analysis" like in the example of the chapter 12 of the Limma guide, but it slighty differs because here I use the same reference in all arrays, so I'm a bit lost on what I should do.

If anyone have an idea on what good design would be good for this analysis it would be very helpful.

Many thanks

This is a standard two colour experiment with a common reference, as covered in Chapter 10 of the limma User's Guide. In fact it's almost identical to the example considered in Section 10.3. You can use

design <- modelMatrix(targets, ref="RNA_reference")

and proceed from there.

There is no need for a single channel analysis, although that also would be possible.