[BioC] limma and paired data

[SKALKO at clinic.ub.es]

Dear Gordon,

Thank you very much for your answer. If I understood well, your
suggestion means to include two commands (shown with **) below?

---
>eset<-rma(x)
>design <- model.matrix(~ -1+factor(c(1,1,1,2,2,2,3,3,4,4,4,5,5,5,6,6)))
>colnames(design) <-
c("group1","group2","group3","group4","group5","group6")

**>cor<-duplicateCorrelation(eset, design,
block=c(1,2,3,4,5,6,7,8,1,2,3,4,5,6,7,8))
**>fit<-lmFit(eset, design, block= c(1,2,3,4,5,6,7,8,1,2,3,4,5,6,7,8),
correlation=cor$consensus)

>contrast.matrix <-
makeContrasts(group2-group1,group4-group1,group5-group2,
group6-group3,levels=design)
>fit2 <- contrasts.fit(fit, contrast.matrix)
>fit2 <- eBayes(fit2)
---

A second question is, what would be the consequences if you do not
calculate "duplicateCorrelation" and use only "block" argument in
"lmFit" (where the default for correlation is 0.75)? That may signify
that I am forcing high correlation between arrays? Is that incorrect, or
a defensible approximation? By the way, I am working with human arrays.

Thank you again for your help and advice

Best,
Susana



-----Mensaje original-----
De: Gordon K Smyth [mailto:smyth at wehi.EDU.AU] 
Enviado el: 23 February 2005 13:58
Para: KALKO, SUSANA (IDIBAPS)
CC: bioconductor at stat.math.ethz.ch
Asunto: [BioC] limma and paired data

Why not follow the section in the limma User's Guide on technical
replication?  Blocking on
individuals and technical replication is really the same thing.

Gordon

> Date: Mon, 21 Feb 2005 11:41:53 +0100
> From: <SKALKO at clinic.ub.es>
> Subject: [BioC] limma and paired data
> To: <bioconductor at stat.math.ethz.ch>
>
> Dear all,
>
>
>
> I have a question on a subject that I think was not discussed here
> before.
>
> I am using limma package for the detection of significant differential
> expression in an affy experiment:
>
>
>
> 3 "Healthy" (group1) indiv. before treatment and the same indiv. after
> treatment (group4)
>
> 3 "ill-low"   (group2)    "                "           "
> "           "                  (group 5)
>
> 2 "ill-high"  (group3)   "                 "           "
> "           "                  (group 6)
>
>
>
> the interest is comparing effects of the treatment (i.e.
group4-group1,
> group5-group2, etc). I used these commands:
>
>
>
>>library(affy)
>
>>library(limma)
>
>>library("hgu133a")
>
>>x<-ReadAffy()
>
>
>
>>eset<-rma(x)
>
>>design <- model.matrix(~
-1+factor(c(1,1,1,2,2,2,3,3,4,4,4,5,5,5,6,6)))
>
>>colnames(design) <-
> c("group1","group2","group3","group4","group5","group6")
>
>
>
>>fit<-lmFit(eset,design)
>
>
>
>>contrast.matrix <-
> makeContrasts(group2-group1,group4-group1,group5-group2,
> group6-group3,levels=design)
>
>
>
>>fit2 <- contrasts.fit(fit, contrast.matrix)
>
>>fit2 <- eBayes(fit2)
>
>
>
> The question is:  How has to be taken into account that the
individuals
> are the same before and after the treatment?
>
> I red about block in lmFit but I am not sure how to do that. Here it
> would be some correlation, but no so high as
>
> in the case of  real technical replicates.
>
>
>
> Thanking you in advance,
>
> Susana Kalko