[BioC] Contrast matrix in limma

[Gordon K Smyth]

> Date: Mon, 14 Feb 2005 12:41:22 +0100
> From: "Ingunn Berget" <ingunn.berget at umb.no>
> Subject: [BioC] Contrast matrix in limma
> To: <bioconductor at stat.math.ethz.ch>
>
> Dear BioC-list
>
> I have data from a microarray experiment comparing 12 different growth conditions for bacteria.
>
> This are the conditions
> BC7ppm BC9ppm CH3COOH EtBr EtOH Glycerol gr15 gr46 HCl NaCl NaOH PK
>
> PK is the normal condition and it is of interest to find genes that are differentially expressed
> in PK and the other conditions. The experiment is a reference design with common reference for all
> arrays, and with three biological replicates from each condition. The reference is always labelled
> Cy3. The two conditions BC7ppm and BC9ppm represent the same medium for the bacteria, but with
> different concentrations of something.
>
> If I have understood the limma usermanual correctly, the following lines will make all contrasts
> for comparing gene epression in one condition with PK such that the contrast names BC7.PK compares
> geneexpression in BC7ppm condition with PK and BC9.PK gene expression in BC9ppm with PPK.
> D <- modelMatrix(MA$target,ref = "Ref") #MA is the MAlist with the normalised data
>
> contrast.matrix <- makeContrasts(gr15.PK=gr15-PK,gr46.PK=gr46-PK,NaCl.PK=NaCl-PK,EtBr.PK=EtBr-PK,
> EtOH.PK=EtOH-PK,NaOH.PK=NaOH-PK,CH2COOH.PK=CH3COOH-PK,HCl.PK=HCl-PK,BC7.PK=BC7ppm-PK,
> BC9.PK=BC9ppm-PK, Gly.PK=Glycerol-PK,levels=D)
>
> Since there is very little difference between BC7ppm and BC9ppm, I want to "merge" these to
> experiments and make one contrast for both. So my questions are
>
> 1) can I do this?
>
> 2) In general the sum of the coefficients in a contrast should be one,

I think you mean, sum to zero.

> does this mean that I
> should use something like
>
> BC.PK= 0.5*BC7ppm+0.5*BC9ppm-PK
>
> as an argument in the makeContrasts command?

Yes, it sounds as if this is what you need.  Your contrast BC.PK compares PK to the average of
BC7ppm and BC9ppm, which sounds like what you're after.

Gordon

> Best regards
>
> Ingunn