William R. Pearson
wrp at dayhoff.med.Virginia.EDU
Thu Apr 28 11:29:01 EST 1994
In article <Coy8nK.DLB at news.otago.ac.nz>,
Craig Marshall <craigm at sanger.otago.ac.nz> wrote:
>std_dickout at hg.uleth.ca wrote:
>> I am interested in information and/or opinions concerning the use of Maximum
>> Parsimony analyses and Distance analyses with respect to the treatment of
>> molecular data in the field of cladistics. Explanations dealing with
>> how these two methods differ from one another would also be appreciated.
>> Please mail me directly.
>There is a very good description of this in Molecular Systematics
>edited by Hillis and Moritz published by Sinauer Associates, where in
>Chapter 11 Swofford and Olsen discuss just this problem. Essentially
>distance bad, parsimony (and others), good.
This is of course, an great over simplification and these
dicussions frequently become religious wars. In a course that I teach
on the subject, students chose protein families and build trees using
the different methods (distance, parsimony, and maximum likelihood).
Considering the fact that I had not pre-selected the families in any
way, I was surprised to learn that with each of the families, it was
possible to get a clearly mistaken tree (plants more closely related
to vertebrates than insects, etc) using one of the methods about 2/3
of the time.
For an alternative view on the appropriateness of parsimony vs
distance methods, see:
%A J. Felsenstein
%T Phylogenies from molecular sequences: Inference and reliability
%J Ann. Rev. Genet.
Swofford may be biased towards parsimony, since he has invested a lot in
the PAUP (Phylogenetic Analysis Using Parsimony) program; Felsenstein's
bias is clearly towards maximum likelihood, which he popularized. Maximum
likelihood has the useful property that it can produce trees that are
very simlar to either parsimony or distance trees, depending on the data.
Distance methods have the great strength that they do not require
multiple alignments; parsimony and maximum likelihood do. The
difficulties encounted in the multiple alignments are often ignored
when comparing the different methods.
I am also a fan of maximum likelihood.
More information about the Bioforum