MrBayes question

Gerald M. Schneeweiss gerald.schneeweiss at UNIVIE.AC.AT
Fri Sep 9 12:11:58 CDT 2005


In the consensus tree of MrBayes, branch lengths calculated as the mean
branch length from all trees in the posterior distribution, in which this
branch is present. If I understood it correctly, in the paper of Lin et al.
branch lengths were optimized on the majority rule consensus tree using the
mean values of the model parameters: this is like optimizing branch lengths
on a given tree in maximum likelihood. During the Markov chain process,
values for all model parameters (which also include the branch lengths) are
proposed, and it can therefore well be that for identical sequences
slightly different branch lengths are proposed and eventually sampled,
resulting in slightly different means of branch lengths in the consensus
tree. A consensus tree with branch lengths from a Bayesian run is fine to
illustrate differences in branch lengths, but I think I would not use it
for other purposes like ancestral character state reconstruction, where
branch lengths are taken into account.

Cheers,
Gerald Schneeweiss




>Dear all,
>I wonder if somebody here could help us with the following
>problem related to the use of the program MrBayes:
>Using command "sumt" with the option "saveblens=yes" we receive
>a consensus tree in the file .con. When this tree is opened in
>TreeView there is a possiblity to get the tree with unequal or equal
>branch lengths. We understand that unequal tree branches reflect
>sequence divergence, but we are not sure exactly in which way this
>divergence is expressed (i.e. how branch lengths are computed) as
>this is in fact consensus tree. In the paper by Lin et al., Syst.
>Biol. 53: 400-421, 2004, the authors use the following formulation:
>"branch lengths are optimised using estimated mean parameter value
>and drawn proportional to character change as indicated by the scale
>bar". We wonder if this is correct and we also wonder how it is possible
>that for identical sequences we obtained slightly different branch
>lengths.
>
>
>* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
>Karol Marhold
>Institute of Botany, Slovak Academy of Sciences
>Dubravska cesta 14, SK-845 23 Bratislava, Slovak Republic
>Tel.: (+421) 2 5942 6128  Fax: (+421) 2 5477 1948
>E-mail: karol.marhold at savba.sk
>http://ibot.sav.sk/Karol01.html
>see also http://www.iopb.org
>and have a look to the prototype of our database
>on chromosome numbers http://147.213.100.121/webapp/
>&
>Department of Botany, Charles University,
>Benatska 2, CZ-128 01 Praha, Czech Republic
>Fax: (+420) 2 2195  1645
>
>* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *




More information about the Taxacom mailing list