[BioC] duplicateCorrelation

[Gordon K Smyth]

Dear Guillaume,

The first approach (using duplicateCorrelation) is correct.  The second 
approach accounts for technical variability, but fails to fully account 
for biological variability of the wt-mu contrast in the standard errors of 
the tests.  The second approach therefore tends to be anti-conservative, 
and will usually lead to more apparent differential expression than the 
first approach.  The second approach is included in the limma User's Guide 
only to give users an alternative when the design is too complex for 
duplicateCorrelation to be applicable.

I am assuming that your experiment is made up of dye-swap technical 
replicates of 3 biological replicates.

Best wishes
Gordon

---------------------------------------------
Professor Gordon K Smyth,
NHMRC Senior Research Fellow,
Bioinformatics Division,
Walter and Eliza Hall Institute of Medical Research,
1G Royal Parade, Parkville, Vic 3052, Australia.
smyth at wehi.edu.au
http://www.wehi.edu.au
http://www.statsci.org/smyth

> Date: Thu, 17 Feb 2011 17:53:04 +0100
> From: Guillaume Meurice <guillaume.meurice at igr.fr>
> To: bioconductor at r-project.org
> Subject: [BioC] duplicateCorrelation
>
> Dear all,
>
>
> I was wondering why there is so many difference between the two 
> following approaches to handle the replication for my experiments.
>
> briefly, Here is my target :
> Cy3	Cy5
> wt1	mu1
> mu1	wt1
> wt2	mu2
> mu2	wt2
> wt3	mu3
> mu3	wt3
>
>
> to get the gene differentially expressed between Mutant and WT, I have 
> stricly followed the two solutions given in the page 37 of limma 
> userguide (3rd apriol 2010):

> - the first one (page 37) is using duplicateCorrelation

> - the second one clearly explicit the design matrix and the contrast 
> matrix (page 38) as follow
>
> design = cbind(
> 		R1_MuvsWT = c(-1,1,0,0,0,0),
> 		R2_MuvsWT = c(0,0,-1,1,0,0),
> 		R3_MuvsWT = c(0,0,0,0,-1,1)
> )
> fit = lmFit(MAn,design)
>
> cont.matrix = makeContrasts (
> 		MuvsWT = (R1_MuvsWT + R2_MUvsWT+R3_MUvsWT)/3,
> 		levels = design
> )
>
>
> using these two approaches give quantitatively different results.
>
>
> Which one should I trust ?
>
> Thanks by advance for any pieces of advice and / or any help
>
> Cheers
>
> --
> G.M

______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}