[BioC] replicate arrays for limma

[Gordon Smyth]

At 06:16 PM 14/12/2003, Simon Melov wrote:
>Hi,
>Im new to limma, and I'm trying to determine a design matrix for the 
>following type of experiment. I dont see an example of this sort of 
>experiment which is becoming increasingly common. I have Diseased vs 
>control (two color). I have 30 diseased individuals, and each individual 
>has had 4-6 technical replicates carried out with dye swaps involved. My 
>question is, how to capitalize on the robustness of the technical reps per 
>individual? Is there a way in limma of obtaining the least variable genes 
>per technical rep set (which I guess violates independence somewhat as the 
>4-6 replicates are done on the same individual), and then comparing these 
>results to all the other 29 diseased individuals (who will have had the 
>same filtering done to identify the most robust differentially expressed 
>genes compared to the control). Ulimatley this will result in the 
>identification of the most robustly differentially expressed genes across 
>all 30 individuals, but will have capitalized on the fact that each 
>individual was technically replicated between 4-6 times.

Is the same control used throughout the experiment? I will assume that it 
is. Here is one way to answer you question. Make up a targets file 
something like this:

Cy3                Cy5
Patient1     Control
Control      Patient1
Patient1     Control
Patient2     Control
Control       Patient2
...

Then in R:

targets <- readTargets()
design <- designMatrix(targets, ref="Control")
fit <- lmFit(MA, design)   # estimate the diseased vs control differences 
for each patient
cont.matrix <- matrix(1,30,1)
fit <- eBayes(contrasts.fit(fit, cont.matrix))  # average the results over 
patients
topTable(fit)


Gordon

>Maybe this is straightforward, but I cant figure out how to do it, please 
>help!
>
>thanks
>
>Simon.