[BioC] Limma factorial experiment: interaction parameter

[Donald Dunbar]

Hello,
I have a gene expression experiment with four replicates each for four 
groups: control, two drugs and their combination.

Control (C.C)
Drug1 (D.C)
Drug2 (C.H)
Drug1+Drug2 (D.H)

I have imported the Affymetrix data, normalised and set up a design to 
analyse in Limma.

    C.C D.C C.H D.H
1    1   0   0   0
2    1   0   0   0
3    1   0   0   0
4    1   0   0   0
5    0   1   0   0
6    0   1   0   0
7    0   1   0   0
8    0   1   0   0
9    0   0   1   0
10   0   0   1   0
11   0   0   1   0
12   0   0   1   0
13   0   0   0   1
14   0   0   0   1
15   0   0   0   1
16   0   0   0   1


I initially set up contrasts and following lmFit, contrasts.fit, eBayes, 
and decideTests find differentially expressed probe-sets:

drug 1 v control contrast 'D.C-C.C' gives 46 probesets
drug 2 v control contrast 'C.H-C.C' gives 233 probesets
drug 1 + drug 2 combination contrast 'D.H-C.C' gives 3897 probesets

Clearly there are several thousand probe-sets differentially expressed 
by the drug combination and not by the individual drugs. I can find 
these crudely by filtering the p-values of the contrasts, but I would 
rather do it properly.

I would like to define an interaction parameter and select 
differentially expressed genes. I read Gordon Smyth's Limma user guide 
and several helpful posts on this list but have so far failed to find an 
appropriate parametrization. The user guide's factorial example has 
sepatare controls for each treatment group so isn't applicable for me I 
think. James MacDonald gave a helpful reply to Suzanne Szak in May 2009 
and I thought I'd try that:

(combo-control)-(x-control)-(y-control) but I think I'm misunderstanding 
his answer: if I try this (D.H-C.C)-(D.C-C.C)-(C.H-C.C) I get no 
significant probe-sets.

My own idea was (D.H-D.C)-(C.H-C.C) but that also gave no significant 
probe-sets and I think was a bit silly.

Any help on this would be greatly appreciated.

With best wishes,
Donald

-- 
Donald Dunbar
University of Edinburgh
donald.dunbar at ed.ac.uk

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: not available
URL: <https://stat.ethz.ch/pipermail/bioconductor/attachments/20110329/5139c235/attachment.pl>