[BioC] Limma analyse factorial data with two-color arrays

James W. MacDonald jmacdon at med.umich.edu
Fri May 22 15:12:58 CEST 2009

Hi Jabez,

Jabez Wilson wrote:
> Sorry.... sent too soon.
> Dear all, I know that this question has been asked in a couple of
> forms, but I haven't noticed a full reply given. I'm hoping that
> someone will be able to give me the exact answer. I'm comparing
> stimulated cells vs unstimulated cells on each slide at two time
> points (4 hrs and 8 hrs). Suppose there are 4 samples at each time
> point the targets file will look like this:
> FileName cy3     cy5 1             stim4  neg4 2             neg4
> stim4 3             stim4  neg4 4             neg4   stim4
> 5             stim8  neg8 6             neg8  stim8 7
> stim8  neg8 8             neg8   stim8
> There is no common reference (pool) as there is in the weaver example
> in the limma guide, so should I use e.g. neg4 as the reference i.e.
> design <- modelMatrix(targets,ref="neg4")
> If I do that then when I fit the model using lmFit(MA, design) I get
> "Coefficients not estimable: stim8 "

You can't use neg4 as a reference when it isn't actually a reference 
(e.g., it has to be on every slide). If you create a design matrix this 
way you will get

 > modelMatrix(targets, ref="neg4")
Found unique target names:
  neg4 neg8 stim4 stim8
      neg8 stim4 stim8
[1,]    0    -1     0
[2,]    0     1     0
[3,]    0    -1     0
[4,]    0     1     0
[5,]    1     0    -1
[6,]   -1     0     1
[7,]    1     0    -1
[8,]   -1     0     1

Which is not of full rank (e.g., the stim8 column is a linear 
combination of the neg8 column).

You don't give any information about your experiment, so it is difficult 
to help. In addition, people are in general hesitant to help people with 
experimental design or analysis questions because a.) that is what many 
of us do for a living, so you are in effect asking for pro bono work, 
and b.) without knowing more about a given experiment it isn't 
reasonable for people to give analysis advice anyway.

So, without knowing more about your experiment other than the short 
names you gave your treatments, can you not simply analyze the '4' 
samples separately from the '8' samples, using a reference design in 
each case? Or if the negX samples are all supposed to be similar in 
expression (and you can show they are), you could rename them 'neg' and 
then have a true reference design.



> Can anyone help me from this point (apart from advising me to do the
> microarray expt with affymetrix chips)?
> [[alternative HTML version deleted]]
> ------------------------------------------------------------------------
> _______________________________________________ Bioconductor mailing
> list Bioconductor at stat.math.ethz.ch 
> https://stat.ethz.ch/mailman/listinfo/bioconductor Search the
> archives:
> http://news.gmane.org/gmane.science.biology.informatics.conductor

James W. MacDonald, M.S.
Douglas Lab
University of Michigan
Department of Human Genetics
5912 Buhl
1241 E. Catherine St.
Ann Arbor MI 48109-5618

More information about the Bioconductor mailing list