In article <6a5c47$t42 at net.bio.net>, Warren Gallin
<wgallin at gpu.srv.ualberta.ca> wrote:
> In Article <6a575g$gpa at net.bio.net>, Joe Staton
<jstaton at oeb.harvard.edu> wrote:
> >The model of evolution is another consideration, altogether (i.e.,
> >parsimony) which MAY be more robust to giving information about
> >evolutionary patterns even if not completely correct. (no flames please)
> To add some data to this discussion, you might want to take a look at Mol.
> Biol. Evol. 14(1):105-108 (1997). "How often do wrong models produce better
> It seems that the maximum likelihood approach is not as sensitive to changes
> in the model as one might guess, a priori. Taking that in conjunction with
> the problem that the assumptions of the maximum parsimony approach often go
> unexamined, I don't think that the rejction of either approach is warranted
> given the current state of knowledge.
I agree. If a difference between an MP tree and an ML tree always leads
to rejecting the ML tree, then what would be the point of building an ML
tree? All of these methods are error-prone and should be considered
> If you get different results with the two approaches, that should lead
> you to examine why that has happened. On the other hand, if you get the
> same results with the two approaches, does that necessarily meant that your
> confidence in the estimated tree should increase?
Definitely not (as I suspect you are implying)! For example, all methods
are vulnerable to the Felsenstein zone (i.e. long branch attraction
[LBA]). I realize that ML can overcome moderately strong cases of LBA if
an appropriate evolutionary model is assumed, but it is certainly not
immune to the problem. My point here is that a matrix that tends to
produce LBA can easily result in the same wrong inference when different
tree estimating methods are applied.
Guy Hoelzer e-mail: hoelzer at med.unr.edu
Department of Biology phone: 702-784-4860
University of Nevada Reno fax: 702-784-1302
Reno, NV 89557