[BioC] [limma] Factorial refernce design

[Gordon K Smyth]

what do you mean by a contrasts matrix "run by lmFit"?  I cannot guess what you can mean.  There
is no contrast argument to lmFit().

You do understand that the limma guide gives three different approaches to analysing the factorial
design, and that your approach is equivalent to the first?  This limma guide section is supposed
to be pedagogic.  You're not supposed to somehow mix all three approaches.

Gordon


On Thu, October 6, 2005 9:05 pm, Ron Ophir wrote:
>
>
>>>> "Gordon K Smyth" <smyth at wehi.EDU.AU> 10/06/05 12:45 PM >>>
>
>> Date: Thu, 06 Oct 2005 12:40:32 +0300
>> From: "Ron Ophir" <ron.ophir at weizmann.ac.il>
>> Subject: [BioC] [limma] Factorial refernce design
>> To: "<bioconcutor" <bioconductor at stat.math.ethz.ch>
>>
>> Hi,
>> I work with dual channel technology and I would like to use limma
> for
>> two way analysis. My experiment is as follow:
>> FileName	                               Cy3	Cy5
>> GL_251322310100_S02_.gpr	WT3h	NI3h
>> GL_251322310101_S02_.gpr	WT3h	NI3h
>> GL_251322310103_S02_.gpr	Mut3h	NI3h
>> GL_251322310104_S02_.gpr	Mut3h	NI3h
>> GL_251322310107_S02_.gpr	WT18h	NI3h
>> GL_251322310136_S02_.gpr	WT18h	NI3h
>> GL_251322310108_S02_.gpr	Mut18h	NI3h
>> GL_251322310368_S02_.gpr	Mut18h	NI3h
>> where I have two factors genotype: WT and Mut and Time: early and
> late
>> and NI3h is the reference.
>> Therefore it seems to be natural that my design matrix would be as
>> 	WT3h	Mut3h	WT18h	Mut18h
>> GL_251322310100_S01	-1	0	0	0
>> GL_251322310101_S01	-1	0	0	0
>> GL_251322310103_S01	0	-1	0	0
>> GL_251322310104_S01	0	-1	0	0
>> GL_251322310107_S01	0	0	-1	0
>> GL_251322310136_S01	0	0	-1	0
>> GL_251322310108_S01	0	0	0	-1
>> GL_251322310368_S01	0	0	0	-1
>>
>> then run fit<-lmFit(Marray,design)
>>
>> I want to answer the following questions:
>> What is the effect of genotype on genes' expression?
>> What is the effect of time on genes' expression?
>> What is the synergistic effect of time and genotype?
>>
>> Since the most intuitive for me is to use the sum to zero
>> parameterization method I created such contrast matrix
>>
>> 	Genotype.WTvsMut	Time.EarlyvsLate	Int
>> WT3h	1	-1	1
>> Mut3h	-1	-1	-1
>> WT18h	1	1	-1
>> Mut18h	-1	1	1
>>
>> then run contrasts<-contrasts.fit(fit,cotrasts.matrix)
>>
>> However from the example in chapter 14 in Limma users guide it seems
>> that I had to run the my contrats matrix with lmFit function.
>
>>I can't guess what you mean, or what difficulty you are imagining.
> The section on factorial
>>designs in the users guide is exactly equivalent to what you have
> given above.
>
>>Gordon
>
> Not exactly, in the user guide the contrasts.matrix as I described here
> is run by lmFit() whereas contrasts.fit() gets
> such matrix
>      WT.SvsU Mu.SvsU Diff
> [1,]       0       0    0
> [2,]       0       0    0
> [3,]      -2      -2    0
> [4,]      -2       2    4
>
> to implement other questions.
>
> I guess that dealing with single channel experiment let run contrasts
> matrix directly in lmFit().
>
> Ron
>
>
>>
>>
>>
>>> I allowed myself to use the example in chapter 14 since reference
>>> design equivalent to affymetrix experiment after parametrizing it
> as
>>> mentioned above with my design matrix and fit it.
>>> So is it right the way I did it although in the example for
> factorial
>>> design my design of  contrasts matrix is run with lmFit function?
>>> Is OK to run it without the intercept WT=c(1,1,1,1) if I'm not
>>> interesting in it?
>>>
>>> Thanks in advance,
>>> Ron
>>
>