[BioC] Limma and time-course data
smyth at wehi.edu.au
Thu Mar 2 02:45:00 CET 2006
You will need at least as many contrasts as time points minus one. So
if you have 3 times, you need at least 2 contrasts. It is just like
At 09:39 AM 2/03/2006, michael watson \(IAH-C\) wrote:
>And a lightbulb goes on just above my head! It was beginning to
>confuse me where the F test came in.
>So once I have fit the contrasts (given that they span the entire
>time course, so if I have endpoint-startpoint as default, I will be
>OK) I can access the F statistic through the fitted model
>object? And this has a significance value associated with it?
>Thanks alot :)
>From: Gordon K Smyth [mailto:smyth at wehi.EDU.AU]
>Sent: Wed 01/03/2006 9:59 PM
>To: michael watson (IAH-C)
>Cc: bioconductor at stat.math.ethz.ch
>Subject: [BioC] Limma and time-course data
>limma does exactly what you want. The approach outlined in the
>User's Guide (and in the workshop)
> finds genes which change over time without worrying about which
> particular time the genes differ
>at. I think you may have missed the fact that the approach uses the
>F-test, not the individual
>contrast p-values. You will get the same F-test regardless of how
>you specify the contrasts, as
>long as the contrasts span all the times.
>As far as verbosity is concerned, limma is a general purpose
>program, not specifically for time
>courses. So to create the F-test, you do need to explicitly setup a
>set of contrasts. Strictly
>speaking, I could get limma to make a set of contrasts automatically
>if it is known that you want
>to do an F-test. But making the contrasts takes only a few lines of
>code (as you show below), so
>I can live with that, at least for now.
> > Date: Tue, 28 Feb 2006 12:30:47 -0000
> > From: "michael watson \(IAH-C\)" <michael.watson at bbsrc.ac.uk>
> > Subject: [BioC] Limma and time-course data
> > To: <bioconductor at stat.math.ethz.ch>
> > Hi
> > Googling the list shows this up to be a rather hot topic, but I just
> > wanted to ask a few more questions.
> > Firstly, it seems the plan for tackling time course data within limma is
> > to treat each time-point/treatment combination as a coefficient to be
> > estimated. Thus, to ask "which genes are changing over time", one must
> > fit contrasts that compare every single time point to every other time
> > point, pairwise, and look for any gene that is significant in one or
> > more of those comparisons. Is that correct?
> > I am also a tad confused by the documentation, which states (on page
> > 47):
> > "> cont.wt <- makeContrasts(
> > + "wt.6hr-wt.0hr",
> > + "wt.24hr-wt.6hr",
> > + levels=design)
> >> fit2 <- contrasts.fit(fit, cont.wt)
> >> fit2 <- eBayes(fit2)
> > Any two contrasts between the three times would give the same result.
> > The same gene list
> > would be obtained had "wt.24hr-wt.0hr" been used in place of
> > "wt.24hr-wt.6hr" for example."
> > I'm confused why "wt.24hr-wt.0hr" and "wt.24hr-wt.6hr" would give the
> > same gene list. The first looks for differences in wt between time
> > points 0 and 24, and the second looks for differences between time
> > points 6 and 24.
> > I guess, to me, this all seems a bit verbose and difficult, particularly
> > for large time-course experiments where many biologists want to subset
> > their data to those genes that change over time and thus want to ask the
> > question "does time have an effect on the expression of my gene?" and
> > are not particularly bothered, at this stage, which particular time
> > points those genes differ at.
> > Thanks in advance
> > Mick
More information about the Bioconductor