[BioC] duplicateCorrelation dye-swap technical replicates

[Gordon Smyth]

Dear Roberta,

I have tried to answer your questions, but nothing that you mention 
is at all surprising or unexpected.

>Date: Thu, 25 Oct 2007 16:04:39 +0200
>From: Roberta Sirovich <roberta.sirovich at unito.it>
>Subject: [BioC] duplicateCorrelation dye-swap technical replicates
>To: bioconductor at stat.math.ethz.ch
>
>Dear all,
>
>I am working with limma package for the analysis of differential
>expression in agilend double color chips.
>My data come from a common reference experiment and the design matrix is
>the following:
>WT   MU
>1   0
>-1   0
>0   1
>0   -1
>1   0
>-1   0
>0   1
>0   -1
>Since rows 1 and 2 are technical dye-swap replicates (as rows 3 and 4, 5
>and 6, 7 and 8), I am trying to use duplicateCorrelation with the block
>argument as follows:
>rep<-(1,1,2,2,3,3,4,4). But there's something that I don't understand.
>1. if I give the following design matrix (instead of the one written above)
>(1,-1,1,-1,1,-1,1-1), the result in corfit$consensus changes (but not
>very much) why?

Well, your original design matrix allows WT and MU to be different, 
while the second design matrix doesn't. If most of the genes are not 
differentially expressed between WT and MU, then the two design 
matrices are equivalent for most genes. Hence consensus results 
across genes are not likely to change much.

>2. if I give as design matix the abs(.) of the matrix above, the result
>in confirt$consensus changes very much. why?

The two design matrices are drastically different, hence give very 
different results. In this case the two design matrices could only be 
equivalent if both WT and MU were the same as the reference, which is unlikely.

>I thought that duplicateCorrelation needs the design matrix to be able
>to distinguish between WT and MU.

Of course.

>  I thought it was't necessary to say
>which chip is dye-swap (it should come from the logratio values?).

Well, you were wrong. Limma needs to know what your design is, 
including the dye-swaps.

>And
>why the result changes if I give the vector in 1.?
>What exactly duplicateCorrelation calculates?

Maybe, read the documentation, or read the cited paper?

Best wishes
Gordon

>Thak you very much!
>Roberta Sirovich
>
>--
>
>-----------------------------
>Dr. Roberta Sirovich
>Department of Mathematics
>University of Torino
>Via Carlo Alberto 10
>10123 Torino, Italy
>tel: +39-011-6702850
>fax: +39-011-6702878