In article <aq6b8p$g8s$1 at mercury.hgmp.mrc.ac.uk>,
Wurff, Andre vander <Andre.vanderWurff at wur.nl> wrote:
>we have used MrBayes software on about 400 full length 18S rDNA (1700 bp)with the TGR model of evolution. With MrBayes, the bootstrap support for deeper (and basically all) phylogenetic nodes is strikingly high (about 100%). Does anybody have the same experience with MrBayes ? Does anybody know some relevant literature on this ? (Our NJ trees with PAUP have remarkably low support at deeper nodes).
>Thanks in advance for any help,
>Andre van der Wurff
This has been widely observed. The reason for it is still
<opinion alert>: I don't believe that the intervals produced by MrBayes
are properly to be compared to bootstrap intervals. Bootstraps ask
"How sensitive are features of my tree to variation in the data?"
Bayesian support intervals ask "What proportion of the nearly-best
trees support my feature?" These are not the same question and will
not necessarily give the same answer.
The example I've used for explaining this is to consider a branch
on a big tree. No sites contradict this branch, and two sites
support it. The Bayesian support may be very strong (depending on
the evolutionary model) because the tree with this branch really
has a much better likelihood than the tree without it. The boostrap
support will probably not be more than 75%, because 1/4 of the
resimulated data sets will lack both key sites.
This would be a theoretical consideration. There is also a practical
one; if MrBayes gets "stuck" in one region of tree space, the error will
always be in the direction of too-tight support intervals. You will
want to look carefully at your runs for signs of non-convergence,
especially with a huge data set like this one. One thing I would
recommend would be asking Paup* to look for "islands" in your
data (or a subset of it, anyway). If there are many islands
of good trees, especially if some of them are hard to find, MrBayes
may have trouble.
The acceptance frequencies of MrBayes, at least the version I used
last year, are not a reliable guide to whether or not it is stuck:
they count all types of moves, and the program may be stuck on its
tree but still moving on its parameters.... It's better to try
multiple runs with different starting points, and look for
consistency of results.
Mary Kuhner mkkuhner at gs.washington.edu