[BioC] HTqPCR and limma and factorial design with no replicates
Naomi Altman
naomi at stat.psu.edu
Sat Jan 30 04:08:56 CET 2010
The reason you have no d.f. for testing is that
you did not fit the model below. You fitted the
model with a separate mean for each treatment
combination, which is equivalent to fitting the
model which includes interaction. So you used up all of your d.f.
Regards,
Naomi
At 11:32 AM 1/29/2010, Andreia Fonseca wrote:
>Dear list,
>
>
>I am going to analyze data with the design shown below. I do not have
>replicates, but is a complete 2x3 factorial design and therefore I should be
>able to fit a linear model
>
>y=mean+Celltype+Treatment+error for each of gene and then do multiple
>testing correction, right? So, why after running,
>
>TS<-paste(raw_files$Cell_Type, raw_files$Treatment, sep=".")
> > TS<-factor(TS, levels=c("NS.control1", "NS.H1",
> > "NS.H2","S.control2","S.H1","
>S.H2"))
> > design<-model.matrix(~0+TS)
> > contrasts<-makeContrasts(TSS.H1-TSS.H2, TSNS.H1-TSNS.H2,
> > (TSNS.H1-TSNS.H2)-(TSS.H1-TSS.H2),levels=design)
> > sr.norm2<-sr.norm[order(featureNames(sr.norm)),]
> > qDE.limma<-limmaCtData(sr.norm2,design=design,contrasts=contrasts,
> > spacing=1)
> >
> > I am getting an error message
> >
> > Error in ebayes(fit = fit, proportion = proportion, stdev.coef.lim =
> > stdev.coef.lim) :
> > No residual degrees of freedom in linear model fits
>
>
>I get this error?
>
>should I only make a two sample t-test? and I can't fit a linear model?
>
>Thanks for your help!
>
>Kind regards
>
>Andreia
>
>
>
>On Thu, Jan 28, 2010 at 9:54 PM, Heidi Dvinge <heidi at ebi.ac.uk> wrote:
>
> > Hi Andreia,
> >
> > > Hello Heidi,
> > >
> > > my question is about the classification of unreliable estimates e
> > > setCategory. I saw the code and it seems that you are estimating just the
> > > classical c.i. based in the variation of the data, right?
> >
> > yep
> >
> > > I have a new
> > > question though concerning qDE.limma, as I told you I have a factorial
> > > design, so I have created a factorial design form the example data from
> > > HT-qPCR
> > >
> > > TSNS.control1 TSNS.H1 TSNS.H2 TSS.control2 TSS.H1 TSS.H2
> > > 1 1 0 0 0 0 0
> > > 2 0 1 0 0 0 0
> > > 3 0 0 1 0 0 0
> > > 4 0 0 0 1 0 0
> > > 5 0 0 0 0 1 0
> > > 6 0 0 0 0 0 1
> > >
> > > TS<-paste(raw_files$Cell_Type, raw_files$Treatment, sep=".")
> > > TS<-factor(TS, levels=c("NS.control1", "NS.H1",
> > > "NS.H2","S.control2","S.H1","S.H2"))
> > > design<-model.matrix(~0+TS)
> > > contrasts<-makeContrasts(TSS.H1-TSS.H2, TSNS.H1-TSNS.H2,
> > > (TSNS.H1-TSNS.H2)-(TSS.H1-TSS.H2),levels=design)
> > > sr.norm2<-sr.norm[order(featureNames(sr.norm)),]
> > > qDE.limma<-limmaCtData(sr.norm2,design=design,contrasts=contrasts,
> > > spacing=1)
> > >
> > > I am getting an error message
> > >
> > > Error in ebayes(fit = fit, proportion = proportion, stdev.coef.lim =
> > > stdev.coef.lim) :
> > > No residual degrees of freedom in linear model fits
> > >
> > >
> > > is it because this that is not adequate for this design? or I have the
> > > wrong
> > > command?
> > >
> > Your approach as such seems valid, but the problem is that you have no
> > replicate arrays. Which means that unfortunately the limma functions won't
> > work, since there's no way to do statistical testing.
> >
> > Maybe someone on the list can help if we know a little more about your
> > arrays/intention. It looks e.g. like you don't use your two control
> > samples at all?
> >
> > Cheers
> > \Heidi
> >
> > > thanks Andreia
> > >
> > >
> > >
> > >
> > > On Wed, Jan 27, 2010 at 11:26 PM, Heidi Dvinge <heidi at ebi.ac.uk> wrote:
> > >
> > >> Hello Andreia,
> > >>
> > >> sorry for the delay in answering. Just to be clear, what confidence
> > >> interval are you referring to? The confidence intervals plotted in
> > >> plotCtOverview? The deviations used to assign categories in then
> > >> filtering
> > >> the data?
> > >>
> > >> Cheers
> > >> \Heidi
> > >>
> > >> P.S. By the way, just in case you're new to R, you can always see the
> > >> source code of a function, by just typing the function name in the
> > >> terminal, without "?" before or "()" after.
> > >>
> > >> > Hi Heidi,
> > >> >
> > >> >
> > >> > sorry is just to say that I have tested to read the file with two
> > >> columns
> > >> > for the two factors and that part works!!! So now I will move on for
> > >> the
> > >> > other functions, plotCtOverview works fine! Now can you just answer to
> > >> the
> > >> > confidence intervals question?
> > >> >
> > >> > Thanks
> > >> > Andreia
> > >> > On Wed, Jan 27, 2010 at 10:24 AM, Andreia Fonseca
> > >> > <andreia.fonseca at gmail.com
> > >> >> wrote:
> > >> >
> > >> >> Hi Heidi,
> > >> >>
> > >> >>
> > >> >> my question is how should be the format of the files.txt file, so
> > >> that
> > >> >> HTqPCR can read in the information of the two different factors,
> > >> should
> > >> >> it
> > >> >> be like the example I wrote below? And what about the confidence
> > >> >> intervals
> > >> >> to filter data, how does the package estimates them?
> > >> >> Cheers
> > >> >> Andreia
> > >> >>
> > >> >>
> > >> >> On Tue, Jan 26, 2010 at 10:23 PM, Heidi Dvinge <heidi at ebi.ac.uk>
> > >> wrote:
> > >> >>
> > >> >>> Hello Andreia,
> > >> >>>
> > >> >>> actually, HTqPCR can handle multi-factor design, there's just no
> > >> >>> example
> > >> >>> of that in the vignette (will consider adding it in the next
> > >> revision).
> > >> >>>
> > >> >>> The function limmaCtData takes all the arguments that you'd use if
> > >> you
> > >> >>> were analysing microarray data using lmFit and contrasts.fit from
> > >> the
> > >> >>> limma package. You need to specify the design and contrast matrix
> > >> >>> yourself though. As I recall, the limma user's guide has a couple of
> > >> >>> example involving factorial design.
> > >> >>>
> > >> >>> HTH
> > >> >>> \Heidi
> > >> >>>
> > >> >>> > Dear list,
> > >> >>> >
> > >> >>> > Soon I will receive data from the qpcr Exicon platform to analyze
> > >> and
> > >> >>> I
> > >> >>> > have
> > >> >>> > been playing around with HTqPCR package, however from the vignete,
> > >> it
> > >> >>> > seems
> > >> >>> > that is only capable of deleting with data design of one factor,
> > >> how
> > >> >>> do
> > >> >>> I
> > >> >>> > handle it with a factorial design, namely how do I read the data
> > >> and
> > >> >>> > create
> > >> >>> > the model.matrix, the examples only consider one factor. Another
> > >> >>> question
> > >> >>> > is
> > >> >>> > concerning part 5 of the vignete, how are the Confidence values
> > >> >>> estimated?
> > >> >>> > is is based in the variance of the data?
> > >> >>> >
> > >> >>> > In order for you to understand my doubts here is my design is a
> > >> 2x3
> > >> >>> > design,
> > >> >>> > I don't have replicates, but each file has the CT values for each
> > >> >>> gene
> > >> >>> > frome
> > >> >>> > pooled RNA of 10 patients.
> > >> >>> > factor1 factor2
> > >> >>> > file1.txt NS C
> > >> >>> > file2.txt NS H1
> > >> >>> > file3.txt NS H2
> > >> >>> > file4.txt S C
> > >> >>> > file5.txt S H1
> > >> >>> > file6.txt S H2
> > >> >>> >
> > >> >>> >
> > >> >>> > Thanks for your help.
> > >> >>> > with kind regards,
> > >> >>> > Andreia
> > >> >>> >
> > >> >>> >
> > >> >>> > --
> > >> >>> > --------------------------------------------
> > >> >>> > Andreia J. Amaral
> > >> >>> > Unidade de Imunologia Clínica
> > >> >>> > Instituto de Medicina Molecular
> > >> >>> > Universidade de Lisboa
> > >> >>> > email: andreiaamaral at fm.ul.pt
> > >> >>> > andreia.fonseca at gmail.com
> > >> >>> >
> > >> >>> > [[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
> > >> >>>
> > >> >>>
> > >> >>>
> > >> >>
> > >> >>
> > >> >> --
> > >> >> --------------------------------------------
> > >> >> Andreia J. Amaral
> > >> >> Unidade de Imunologia Clínica
> > >> >> Instituto de Medicina Molecular
> > >> >> Universidade de Lisboa
> > >> >> email: andreiaamaral at fm.ul.pt
> > >> >> andreia.fonseca at gmail.com
> > >> >>
> > >> >
> > >> >
> > >> >
> > >> > --
> > >> > --------------------------------------------
> > >> > Andreia J. Amaral
> > >> > Unidade de Imunologia Clínica
> > >> > Instituto de Medicina Molecular
> > >> > Universidade de Lisboa
> > >> > email: andreiaamaral at fm.ul.pt
> > >> > andreia.fonseca at gmail.com
> > >> >
> > >>
> > >>
> > >>
> > >
> > >
> > > --
> > > --------------------------------------------
> > > Andreia J. Amaral
> > > Unidade de Imunologia Clínica
> > > Instituto de Medicina Molecular
> > > Universidade de Lisboa
> > > email: andreiaamaral at fm.ul.pt
> > > andreia.fonseca at gmail.com
> > >
> >
> >
> >
>
>
>--
>--------------------------------------------
>Andreia J. Amaral
>Unidade de Imunologia Clínica
>Instituto de Medicina Molecular
>Universidade de Lisboa
>email: andreiaamaral at fm.ul.pt
> andreia.fonseca at gmail.com
>
> [[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
Naomi S. Altman 814-865-3791 (voice)
Associate Professor
Dept. of Statistics 814-863-7114 (fax)
Penn State University 814-865-1348 (Statistics)
University Park, PA 16802-2111
More information about the Bioconductor
mailing list