[BioC] self-self hybridization and limma

[Sean Davis]

On Mar 29, 2005, at 2:46 PM, Na, Ren wrote:

> hello,
>
> If we have many samples to be compared in an microarray experiment,
> for example,
>
> 	tissueType1	tissueType2	tissueType3
> age1	4		4		4
> age2	4		4		4
> age3	4		4		4
> each kind of sample has 4 biological replicates, primary interest are 
> differential
> expression among different age groups and among different tissueTypes. 
> We usually
> use common reference design. I am wondering if I can use self-self 
> hybridization design,
> in which two identical samples are labeled with different dyes and 
> hybridized to the
> same slide. maybe I don't need to worry about dye bias by using 
> log-intensity A-value
> for each spot, and use limma analyze like,
> MA<-normalizeWithinArrays(RG, method="none")
> MA<-normalizeBetweenArrays(MA, method="Aq")
> convert MA to exprSet, then replace M-value in exprSet with A-value, 
> then use the new
> exprSet to get significant genes using limma. I only know self-self 
> experiment to be
> used to show imbalance in red and green intensity, but I never found 
> it to be used to
> do real experiment. I think there must be some reasons that self-self 
> hybridization is
> not appropriate.
> Could anyone explain it, Thanks in advance!
>

These are some useful links for thinking about factorial designs.  Note 
that the limma user guide also contains examples of factorial design.

http://www.bioconductor.org/workshops/Heidelberg02/exp-design.pdf
http://www.maths.adelaide.edu.au/people/psolomon/Designsingle.pdf
http://www.microarrays.med.uni-goettingen.de/landgrebe_et_al2004b.pdf

In practice, a direct design can cut the variance in half when 
comparing two samples on two arrays via a common reference versus the 
same two samples on a single array, so they can be very useful.  The 
dye bias is a real phenomenon, so needs to be accounted for in the 
design of the experiment (dye swaps).

Sean