That looks right.
The correlation that is computed is the within subject
correlation. In the limma "block" model, the correlation is the same
for 0-6, 6-24 and 0-24 hours. Although this is probably a
simplification, when there are only 3 time points and the pairwise
comparisons are of most interest the tests obtained from this simple
correlation structure do not differ much (usually) from most
complicated modeling of the correlation.
--Naomi
At 12:18 PM 10/24/2007, Mayra Eduardoff wrote:
>Dear Naomi,
>
>Thanks for your reply, can I then do it like this somehow:
>
>data is my expressionSet containing all 39 Arrays, first 0h then 6h
>then 24h always the same order of samples.
>
> > block <- rep(1:13,3)
> > design <- cbind(c(rep(1,13),rep(0,26)),
> c(rep(0,13),rep(1,13),rep(0,13)), > c(rep(0,26),rep(1,13)))
> > rownames(design) <- c("0h","6h","24")
>
>I ram not quite sure what this line does, as I don't understand what
>the correlation parameter expects in my case, but it comes from an
>example.
>
>dupcor <- duplicateCorrelation(exprs(data),design,block=block)
>
>
> > fit <- lmFit(exprs(data),design=design, block=block, correlation=
> dupcor$consensus)
> > cont.wt <- makeContrasts("6h-0h","24h-0h", levels=design)
> > fit2<- contrasts.fit(fit,cont.wt)
>
>
>
>kind regards,
>Mayra
>
>
>
>
>On 10/24/07, Naomi Altman <naomi at="" stat.psu.edu=""> wrote:
> > Dear Mayra,
> > The variance estimator comes from doing an ANOVA with patient as
> > block. The t-tests you want are then done using contrasts.
> >
> > --Naomi
> >
> > At 04:24 AM 10/24/2007, Mayra Eduardoff wrote:
> > >Hi,
> > >I am having problems fitting the following model into limma, I
have
> > >tried many different ways of specifying the design and contrast
matrix
> > >but they all seem nonsense so I won t post them here, but maybe
anyone
> > >knows how to properly fit this:
> > >
> > >I have data from 13 Patients, sampled at three different time
points,
> > >0h (Control), 6h Treatment, 24h Treatment, so all in all 39
arrays and
> > >I want to do paired t tests for differential expression between
6h-0h
> > >and 24h-0h (not ! an ANOVA over both Time points). Until now we
have
> > >been doing this fitting the model for the 6h and 24h separately,
which
> > >works fine, but in oder to take all data into account for the
mod. t
> > >tests variance estimator I guess i need to fit all data into one
> > >model.
> > >
> > >kind regards,
> > >
> > >Mayra
> > >
> > >_______________________________________________
> > >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
> >
> >
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
