[BioC] limma_design_matrix.

[Gordon Smyth]

At 11:09 PM 30/06/2004, Marcus Gry Björklund wrote:
>Hi.
>
>I´m trying to look for differentially expressed genes using a saturated
>design. I have 5 samples which are hybridized in the following scheme.
>
>Sample 1 -> sample2, sample3 (sample 1 is hybridized to sample 2 and
>sample3, sample one is colored by cy5 in both hybridizations and sample 2
>and 3 are labelled with cy3)
>
>Sample 2 -> sample 3, sample 4
>Sample 3 -> sample 4, sample5
>Sample 4 -> sample 1, sample 5
>Sample 5 -> sample1, sample2
>
>It is easier to visualize if you draw it on a paper, it will look like a
>pentagram-like structure.
>
>The samples are RNA from two cell lines which has been mixed in different
>proportions, and I would like to see which genes pops up as differentially
>expressed. I have used the Maanova package from Gary churchill, and I would
>like to compare the result using the Btest.
>
>Do someone know how the design-matrix should be done?

One way would be to create a 'targets' data.frame, as in section 7.3 of the 
Limma User's Guide, and to use

design <- modelMatrix(targets, ref="Sample 1")

This would create 4 coefficients comparing each of the other samples to 
Sample 1.

Gordon

>Best regards
>/Marcus