[BioC] LIMMA:design and contrast matrices for biological and technical replicates

Mike White mikewhite at drexel.edu
Thu Apr 13 21:56:36 CEST 2006


Adai,

Thanks for the quick response. As I said in the original posting I am  
new to this area, and so some of my questions/concepts may be naive  
or incorrect.

Given the nature of the experiment (mice are either treated with  
nicotine ("nic") or not ("contr")) and we want to examine genes up or  
down-regulated by nicotine exposure, I don't see how the "nic" would  
be confounded with "contr". Of course, I'm not a statistics maven,  
and I may have misunderstood what you were getting at.

File1 and File2 should be biologicaly identical, as the probes were  
prepared from the same RNA preparations and should be technical  
replicates for those respective RNA preparations. File3 and File4 are  
technical replicates from RNA isolated from a different pair of  
animals, and File5 and File6 are technical replicates from yet  
another set of animals.

contr1, contr2, and contr3 are biological replicates from three  
different control animals, and nic1, nic2, and nic3 are biological  
replicates from three different nicotine-teated animals. As they are  
biological replicates, I don't think that the three contr samples  
should be a priori identical, but should be (hopefully) very similar;  
the same should hold for the three nic samples.

I think that the first method I tried in the posting (the one that  
worked) is sufficient for this type of experiment, but I wanted to  
try the other way, as it seemed more general and adaptable to more  
complex situations that may come up in the future. In other words,  
I'm trying to climb learning curve so that I can use the technology  
and analysis methods to its fullest.

Mike




On Apr 12, 2006, at 6:54 PM, Adaikalavan Ramasamy wrote:

> Your "nic" variable is confounded with "contr" variable, therefore not
> estimable.
>
> Can you clarify the following please :
>  1. Do you expect File1 and File2 to be biologically identical ?
>  2. Do you expect contr1, contr2, contr3 to be identical (i.e. your  
> used
> a universal RNA pool for all six arrays) ?
>
> If the answer is yes to both of the above, then you might need  
> something
> along the lines of
>
> samples <- as.factor( c("nic1", "nic1", "nic2", "nic2", "nic3",  
> "nic3"))
> design  <- model.matrix( ~ -1 + samples )
>
>   samplesnic1 samplesnic2 samplesnic3
> 1           1           0           0
> 2           1           0           0
> 3           0           1           0
> 4           0           1           0
> 5           0           0           1
> 6           0           0           1
>
> Regards, Adai
>
>
> On Wed, 2006-04-12 at 12:12 -0400, Mike White wrote:
>> Hello,
>>
>> 	I have started using limma to analyze data obtained from experiments
>> examining the effects of nicotine exposure on gene expression in
>> defined regions of the central nervous system. I am using R v2.2.1
>> and limma v2.4.13 running under linux. The data are obtained using 2-
>> color microarrays using probes made from three different mice and
>> duplicate arrays for each set of probes, giving 6 arrays (3
>> biological and two replicates):
>>
>>         Cy3     Cy5
>> File1    contr1 nic1
>> File2    contr1 nic1
>> File3    contr2 nic2
>> File4    contr2 nic2
>> File5   contr3 nic3
>> File6   contr3 nic3
>>
>> I have tried several different ways of setting up the design and
>> linear model to fit, including one similar to the one suggested by
>> Gordon in his posting of 28 Sept 05:
>>
>> design<-cbind(nic1vscontr1=c(1,1,0,0,0,0), nic2vscontr2=c
>> (0,0,1,1,0,0), nic3vscontr3=c(0,0,0,0,1,1))
>> cont.matrix<- makeContrasts(nicvscontr= c(1,1,1)/3, levels=design)
>>
>> this does return  results (of course, how meaningful they are
>> requires more work...).
>>
>> However, I also tried an alternate way of setting things up following
>> an example in section 23.5 ("Technical replication") in Gordon's
>> chapter in the Bioconductor book in which one of the controls is set
>> as a reference and everything is done in relation to this. That
>> particular example represented a more complex situation than the one
>> here, but I wanted to see how this compared to the other method and
>> assumed that it could be applied to my situation.:
>>
>> design<- modelMatrix(targets, ref= "contr1")
>> Found unique target names:
>>   nic1  nic2  nic3  contr1 contr2 contr3
>> colnames(design)
>>
>> [1] "nic1" "nic2" "nic3" "contr2" "contr3"
>>
>>
>>
>> the design matrix is as follows
>>
>>                          nic1   nic2   nic3    contr2      contr3
>> File1                  1       0           0          
>> 0                0
>> FIle2                  1       0           0           
>> 0               0
>> File3                 0        1           0          
>> -1              0
>> File4                 0       1           0           
>> -1              0
>> File5                 0         0         1            
>> 0              -1
>> File6                 0         0         1            
>> 0             -1
>>
>> which is what I expected
>>
>>
>> however when I try to fit the data the following happens
>>
>> fit<- lmFit(MA,design)
>> Coefficients not estimable: contr2 contr3
>>
>> I am a neophyte with both microarrays and limma, and am still feeling
>> my way around setting up design and contrast matrices. However, I
>> can't understand why the second method fails. Any insights?
>>
>> Thanks
>>
>> Mike White
>>
>> ----------------------------------------------------------
>> Michael M. White, Ph.D.
>> Department of Pharmacology & Physiology
>> MS #488
>> Drexel University College of Medicine
>> 245 N. 15th Street
>> Philadelphia, PA 19102-1192
>>
>> phone: 215-762-2355
>> fax: 215-762-4850
>>
>>
>>
>> 	[[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
>>
>



More information about the Bioconductor mailing list