[BioC] design a modelMatrix with no common references

[Robert Castelo]

hi Giusy,

so then I understand that the only way to go with two-color 2x2
factorial design without a common reference RNA source is to treat each
combination of factors as an independent experiment, right?

(i initially understood from the last emails that there was an
alternative by using the parameters argument to the modelMatrix()
function, but i understand now this alternative cannot be employed)

robert.

On Mon, 2009-07-13 at 11:23 -0400, Giusy Della Gatta wrote:
> Hi Robert,
> I saw the example used in chapter  11.5. The problem is that
> I don't have any common reference that I can use to normalize
> the data, while in the chapter they are a comparing everything against
> the "Pool".
> 
> 
> G
> 
> 
> 
> 
> -----Original Message-----
> From: Robert Castelo [mailto:robert.castelo at upf.edu] 
> Sent: Monday, July 13, 2009 6:00 AM
> To: James W. MacDonald
> Cc: Giusy Della Gatta; bioconductor at stat.math.ethz.ch
> Subject: Re: [BioC] design a modelMatrix with no common references
> 
> James, Guisy,
> 
> if you let me make a question about what you're discussing.
> 
> why do you say that the two-color 2x2 factorial design without a common
> reference RNA source is "a lot of work" ??
> 
> (i was actually wondering how would be the 11.5 example of the Limma
> user's guide without a common RNA source)
> 
> thanks!
> robert.
> 
> On Fri, 2009-07-10 at 10:14 -0400, James W. MacDonald wrote:
> > Hi Guisy,
> > 
> > You really have two different experiments here, so I don't know if limma 
> > is going to want to do things automatically for you without warnings or 
> > incorrect model matrices. However, I think you want to use the 
> > parameters argument to modelMatrix() rather than the ref argument (since 
> > you have two different reference samples).
> > 
> >  > targets <- matrix(paste(rep(c("Myc","Rag"), each=4), 
> > rep(c("CD3","PBS"), each=2, times=3)[2:9], sep = "_"), byrow=T, ncol=2)
> >  > targets
> >       [,1]      [,2]
> > [1,] "Myc_CD3" "Myc_PBS"
> > [2,] "Myc_PBS" "Myc_CD3"
> > [3,] "Rag_CD3" "Rag_PBS"
> > [4,] "Rag_PBS" "Rag_CD3"
> >  > colnames(targets) <- c("Cy3","Cy5")
> >  > rownames(targets) <- paste("Array", 1:4)
> >  > targets
> >          Cy3       Cy5
> > Array 1 "Myc_CD3" "Myc_PBS"
> > Array 2 "Myc_PBS" "Myc_CD3"
> > Array 3 "Rag_CD3" "Rag_PBS"
> > Array 4 "Rag_PBS" "Rag_CD3"
> >  > parameters <- cbind(First=c(-1,1,0,0), Second=c(0,0,-1,1))
> >  > rownames(parameters) <- c("Myc_PBS","Myc_CD3","Rag_PBS","Rag_CD3")
> >  > parameters
> >          First Second
> > Myc_PBS    -1      0
> > Myc_CD3     1      0
> > Rag_PBS     0     -1
> > Rag_CD3     0      1
> >  > modelMatrix(targets, parameters)
> > Found unique target names:
> >   Myc_CD3 Myc_PBS Rag_CD3 Rag_PBS
> >          First Second
> > Array 1    -1      0
> > Array 2     1      0
> > Array 3     0     -1
> > Array 4     0      1
> > Warning message:
> > In modelMatrix(targets, parameters) :
> >    number of parameters should be one less than number of targets
> > 
> > But that seems like a lot of work, as the parameters matrix is exactly 
> > the model matrix you want.
> > 
> > Best
> > 
> > Giusy Della Gatta wrote:
> > > Hi everybody,
> > > 
> > > I have Agilent two colors expression arrays in which have been analyzed 
> > > two KO mice samples (myc-/- and Rag -/-) treated with CD3 and with PBS.
> > > I have a total of 4 arrays composed as follows:
> > >      Sample   Cy3         Cy5
> > > 1. Myc24CD3	Myc_CD3	Myc_PBS (Swap)	
> > > 2. Myc24PBS	Myc_PBS	Myc_CD3	
> > > 3. Rag24CD3	Rag_CD3	Rag_PBS (Swap)		
> > > 4. Rag24PBS	Rag_PBS	Rag_CD3	
> > > 
> > > After the normalization I don't know
> > > how to proceed for the  construction of the model matrix.
> > > 
> > > By using the suggestions of the  "Direct Two Color Designs" example (chapter 7.4 LIMMA guide)
> > > I did:
> > > 
> > > 
> > >> targets
> > >                  FileName     Cy3     Cy5 Collection_time
> > > 1 3_Myc24CD3gr_Myc24PBSre Myc_CD3 Myc_PBS             24h
> > > 2 9_Myc24PBSgr_Myc24CD3re Myc_PBS Myc_CD3             24h
> > > 3 5_Rag24CD3gr_Rag24PBSre Rag_CD3 Rag_PBS             24h
> > > 4 4_Rag24PBSgr_Rag24CD3re Rag_PBS Rag_CD3             24h
> > > 
> > >> designmyc= modelMatrix(targets, ref="Myc_PBS")
> > > Found unique target names:
> > >  Myc_CD3 Myc_PBS Rag_CD3 Rag_PBS
> > > 
> > >> designmyc
> > >      Myc_CD3 Rag_CD3 Rag_PBS
> > > [1,]      -1       0       0
> > > [2,]       1       0       0
> > > [3,]       0      -1       1
> > > [4,]       0       1      -1
> > > 
> > >> fit = lmFit(MA.Rq, designmyc)
> > > Coefficients not estimable: Rag_PBS 
> > > Warning message:
> > > Partial NA coefficients for 45018 probe(s)
> > > 
> > > 
> > > But at this point I calculated just the ratios of Myc_CD3/Myc_PBS
> > > and Rag_Myc/Myc_PBS (I am not really interested in this last one!).
> > > How can I specify in the model matrix design that I need two different references
> > > to calculate the following logratios:  Myc_CD3/Myc_PBS, Rag_Myc/Rag_PBS?
> > > 
> > > 
> > > Thank you in advance!
> > > Giusy
> > >  
> > > 
> > > _______________________________________________
> > > 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
> > 
> 
>