[BioC] limma - design matrix for time series

[Sebastian Mueller]

dear members,

I have a general question about the creation of design matrices for 
non-reference designs in limma.

The set-up of the experiment I'm refering to looks like this:
------------------
 0h		1h		2h		4h	

 	3		7		11
 A	>	C	>	E	>	G
 ^ 1		^ 5		^ 9		^ 13
 v 2		v 6		v 10		v 14
 B	>	D	>	F	>	H
 	4		8		12
------------------
The experiment consists, as you hopefully can guess, of 14 arrays (each arrow 
is a two-color array which points from cyc5 to cyc3).
Array 1 and 2 (5-6,9-10,13-14) correspond to a dye-swaps.
Each letter (A-H) corresponds to another condition (however A and B should be 
the same condition)

So, I'm interrested in the contrasts A-B, C-D, E-F, G-H for the time points 
0h, 1h, 2h, 4h, respectively.

The easiest way to do this (from the limma guide), would probably be to create 
a target-frame...

targets <- as.data.frame(cbind(	
Cy3=c("B","A","A","B","D","C","C","D","F","E","E","F","H","G"),
Cy5=c("A","B","C","D","C","D","E","F","E","F","G","H","G","H")))
rownames(targets) <- paste("Array",1:14)
> targets
         Cy3 Cy5
Array 1    B   A
Array 2    A   B
Array 3    A   C
Array 4    B   D
Array 5    D   C
Array 6    C   D
Array 7    C   E
Array 8    D   F
Array 9    F   E
Array 10   E   F
Array 11   E   G
Array 12   F   H
Array 13   H   G
Array 14   G   H

... and creating a design-matrix:
 
> design=modelMatrix(targets,ref="A")
Found unique target names:
 A B C D E F G H
> design
              B  C  D  E  F  G  H
Array 1  -1  0  0  0  0  0  0
Array 2   1  0  0  0  0  0  0
Array 3   0  1  0  0  0  0  0
Array 4  -1  0  1  0  0  0  0
Array 5   0  1 -1  0  0  0  0
Array 6   0 -1  1  0  0  0  0
Array 7   0 -1  0  1  0  0  0
Array 8   0  0 -1  0  1  0  0
Array 9   0  0  0  1 -1  0  0
Array 10  0  0  0 -1  1  0  0
Array 11  0  0  0 -1  0  1  0
Array 12  0  0  0  0 -1  0  1
Array 13  0  0  0  0  0  1 -1
Array 14  0  0  0  0  0 -1  1

cont=makeContrasts(C-D,E-F,G-H,levels=design)
...

---------------------
As I actually don't have a reference design, I'm wondering if this is ok to do 
this. 
My problem is that I can't get the contrast A-B from this design. I already 
tried to append a colum A to this matrix by hand, but calling 
>lmFit(MA, design)
gives me: Coefficients not estimable: A

As I probably haven't really understood the concept of creating a design 
matrix (I have already looked trough the limma guide), I was wondering if 
someone knows a good tutorial or book (I was already looking for hours, but 
there is not such a thing for limma and time series analysis)

I'm espacially wondering how I can solve this problem intuitively (as I was 
going to use limma for my future time-series experiments) and I also would 
like to know if my approch to tackle this problem is appropriate in general.

Thanks a lot

Sebastian Mueller

------
Max-Planck-Institute for Chemical Ecology
Department of Biochemistry
Hans-Knoell-Strasse 8 
D-07745 Jena 
Germany