[BioC] Paired two-color design

[Gordon K Smyth]

Dear January,

Are your dye-swaps technical replicates of the same RNA samples?  That is, 
does "A1" refer to same RNA sample for the second array as does for the 
first?

lmscFit() can't be used in conjunction with technical replicates, 
including dye-swaps.

On the other hand, your paired design is exactly of the form discussed on 
page 41 of the limma User's Guide (dated 10 June 2012):

http://bioconductor.org/packages/2.10/bioc/vignettes/limma/inst/doc/usersguide.pdf

or on page 56 of the limma User's Guide dated 31 July 2012:

http://bioconductor.org/packages/2.11/bioc/vignettes/limma/inst/doc/usersguide.pdf

Best wishes
Gordon

> Date: Tue, 28 Aug 2012 08:17:39 +0200
> From: January Weiner <january.weiner at gmail.com>
> To: bioconductor at r-project.org
> Subject: [BioC] Paired two-color design
>
> Hello,
>
> I have a problem with the design for an experiment that I'm evaluating.
>
> There are two parts to that problem.
>
> The first part: we have two groups (A and B) after treatment, each in
> three replicate, and two-color arrays have been created, with a
> dye-swap for each replicate:
>
> Cy3 Cy5
> A1  B1
> B1  A1
> A2  B2
> B2  A2
> A3  B3
> B3  A3
>
> Here is how I'm doing it currently, but I think that this might not be
> the optimal solution:
>
> design <- cbind( ex1 = c( -1, 1, 1, -1, 0, 0, 0, 0, 0, 0, 0, 0 ), ex2
> = c( 0, 0, 0, 0, -1, 1, 1, -1, 0, 0, 0, 0 ), ex3 = c( 0, 0, 0, 0, 0,
> 0, 0, 0, -1, 1, 1, -1 ) )
> corfit <- intraspotCorrelation( rg, design ) # rg contains the normalized arrays
> fit <- lmscFit( rg, design, correlation= corfit$consensus.correlation )
> cmtx <- makeContrasts( "(ex1 + ex2 + ex3)/3", levels= design )
> fit <- contrasts.fit( fit, cmtx )
> fit <- eBayes( fit )
>
> Is this correct? Do I have to use intraspotCorrelation to detach the
> arrays? I think this should work somehow directly, but I can't figure
> out how.
>
> The second problem is this: we have the same setup as above, except
> that there are controls for A and B (before treatment). I call them
> ctrlA and ctrlB, respectively. There is only one biological replicate
> with a dye swap for each of these controls. The setup looks now like
> that:
>
> targets:
> Cy3 Cy5
> ctrlA ctrlB
> ctrlB ctrlA
> A1  B1
> B1  A1
> A2  B2
> B2  A2
> A3  B3
> B3  A3
>
> I want to look for interaction between the treatment effect and the
> group, in other words I want the difference between (A vs controlA) -
> (B vs controlB).
>
> I use more or less the same approach as above (using intraspotCorrelation):
>
> t2 <- targetsA2C( t )
> design <- model.matrix( ~ 0 + t$group )
> colnames( design ) <- levels( t$group )
> # colnames( design ) are: ctrlA, ctrlB, A1, A2, A3, B1, B2, B3
>
> corfit <- intraSpotCorrelation( rg, design )
> fit <- lmsciFit( rg, design, correlation= corfit$consensus.correlation )
> cmtx <- makeContrasts( "(( A1 + A2 + A3 )/3 - ctrlA) - ( B1 + B2 + B3
> )/3 - ctrlB))", levels= design )
> fit <- contrasts.fit( fit, cmtx )
> fit <- eBayes( fit )
>
> Is that correct?
>
> Kind regards,
>
> J.
>
> -- 
> -------- Dr. January Weiner 3 --------------------------------------
>

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