Hi Sorry, it's been a while since I worked with a common reference design, so I may be repeating myself. I have some data, two conditions both compared to a common reference (which is always Cy3). So targets looks something like this: Array Cy3 Cy5 1 ref T1 2 ref T1 3 ref T1 4 ref T2 5 ref T2 6 ref T2 What I want to do is find those genes which are differentially expressed between T1 and T2 (in either direction; instinctively I want to use a two-sample t-test). So I have a design matrix so: T1 T2 1 0 1 2 0 1 3 0 1 4 1 0 5 1 0 6 1 0 And a contrasts matrix so: T1 T2 T1-T2 T1 1 0 1 T2 0 1 -1 Is this right? From reading the examples, I think it is, but I am not sure what the advantage is of having the contrasts matrix as well as the design matrix, adding the contrasts matrix certainly gives me different results to just using the design matrix alone. Hmmmmm. Thanks for any help! Mick

On Oct 8, 2004, at 10:57 AM, michael watson (IAH-C) wrote: > > Is this right? From reading the examples, I think it is, but I am not > sure what the advantage is of having the contrasts matrix as well as > the > design matrix, adding the contrasts matrix certainly gives me different > results to just using the design matrix alone. Hmmmmm. > Mick, The answer depends on what you want to get? I am assuming that you actually want to have T1-T2, so you do need the contrast matrix. You don't get that quantity directly from the design matrix, so there is no comparable answer using a design matrix alone as compared to including a contrast matrix. The first two coefficients for each should be the same, though. Sean

Hi Sean I want to do in limma what I would have normally done using mt.teststat in the multtest package - that is perform a moderated t-test for each row of my data, testing for the difference between the two groups (where one group is T1/Ref and the other is T2/Ref), adjusting for the FDR. It looks like I do need the contrasts matrix, and the coefficient of interest is T1-T2. So If I use design matrix T1 T2 1 0 1 2 0 1 3 0 1 4 1 0 5 1 0 6 1 0 And a contrasts matrix so: T1 T2 T1-T2 T1 1 0 1 T2 0 1 -1 Then lmfit(), eBayes and topTable, as in the user guide, that's the limma equivalent of a moderated t-test for each row as described above? Cheers Mick -----Original Message----- From: Sean Davis [mailto:sdavis2@mail.nih.gov] Sent: 08 October 2004 17:31 To: michael watson (IAH-C) Cc: Bioconductor Subject: Re: [BioC] Limma with common Reference Design On Oct 8, 2004, at 10:57 AM, michael watson (IAH-C) wrote: > > Is this right? From reading the examples, I think it is, but I am not > sure what the advantage is of having the contrasts matrix as well as > the design matrix, adding the contrasts matrix certainly gives me > different results to just using the design matrix alone. Hmmmmm. > Mick, The answer depends on what you want to get? I am assuming that you actually want to have T1-T2, so you do need the contrast matrix. You don't get that quantity directly from the design matrix, so there is no comparable answer using a design matrix alone as compared to including a contrast matrix. The first two coefficients for each should be the same, though. Sean