Hi Scott,
The reason for implementing only the consensus on the topology was after
reading the appropriate chapter in Inferring Phylogenies. So I'm glad
that Joe himself stepped in. I have added this reference in the help
page (it was already cited in my book with respect to this issue).
I think that the usage of branch lengths on (or from) a consensus tree
depends on your perspective on phylogenetic estimation.
If you are a frequentist, the goal of resampling the data is to give an
idea of the accuracy of the estimates. So you might not use the
bootstrap trees to estimate branch lengths; you already did that by ML
(or maybe least squares) with the full data.
If you are Bayesian, the trees sampled from an MCMC are here for
estimation including of the branch lengths, so you use them to compute
some sort of consensus topology as well as its branch lengths. So it
makes sense that MrBayes can do a consensus tree with branch lengths.
Cheers,
Emmanuel
Scott Chamberlain wrote on 22/11/2012 09:41:
Dear Joe,
Thanks for your feedback on this question. I will go read those pages you
mentioned.
Scott
On Wed, Nov 21, 2012 at 5:19 PM, Joe Felsensteinj...@gs.washington.eduwrote:
Daniel Barker wrote:
What should branch lengths on a consensus tree represent?
Scott Chamberlain had written:
When making a consensus tree using ape::consensus the branch lengths are
lost. Is there a way to not lose the branch lengths? Or to add them
somehow
to the consensus tree after making it.
The issue of what branch lengths ought to be on a consensus tree is not
simple. If we have three rooted trees:
((A:1,B:1):1,C:2);
((A:1,B:1):1,C:2);
(A:2,(B:1,C:1):1);
the consensus tree should be the first tree, but what branch length should
be used for (say) the branch ancestral to the AB clade? 1? 0.667?
The minute you open this can of worms it becomes clear that the answer
depends on what you want that number to convey and what interpretations
your audience will tend to draw from the number. There is no obvious
answer. So this is not a mere technical computing question.
By the way, in my 2004 book, you will find me agonizing about this on page
526, coming down on the side of 0.667, but not overwhelmingly convincingly.
You could argue that a branch length should be set 0 when the branch is
not there, and all the resulting values averaged, or you could argue that
the average should only be taken over those trees for which that branch is
present.
One possible way to solve the problem is to take the consensus tree as if
it were a user-defined tree, use your whole data set, and infer branch
lengths on that tree. Daniel has already expressed his legitimate concerns
in such a case as to whether it takes (for example) trifurcations as if
they were real rather than an expression of our uncertainty.
J.F.
Joe Felsenstein j...@gs.washington.edu
Department of Genome Sciences and Department of Biology,
University of Washington, Box 355065, Seattle, WA 98195-5065 USA
(from 1 October 2012 to 10 December 2012 on sabbatical leave at)
Department of Statistics, University of California, Berkeley, 367 Evans
Hall, Berkeley, CA 94710
[[alternative HTML version deleted]]
___
R-sig-phylo mailing list
R-sig-phylo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
--
Emmanuel Paradis
IRD, Jakarta, Indonesia
http://ape.mpl.ird.fr/
___
R-sig-phylo mailing list
R-sig-phylo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
