Quartet puzzling

Byron Adams bjadams at biocomp.unl.edu
Wed Jan 7 19:56:47 EST 1998

>Has anybody analysed any datasets using 'regular' maximum likelihood methods
>and also with Quartet Puzzling?  Are there instances when the two can give
>very different results? If so, what might those instances be?

Hi James -

	I've played around with these algorithms a little bit using
different data sets, and I've found that quartet puzzling often will not
give the "best" or most likely topological arrangement found using other
likelihood analyses, such as DNAML.  However, in my experience this
different arrangement has never been shown to be "significantly different"
by a log-likelihood test.  I discussed this with K. Strimmer, the author of
PUZZLE (a program that uses quartet puzzling), who suggested that
differences in the tree search algorithm, specifically stepwise addition
and branchswapping (as implemented in DNAML, versus quartet puzzling) are
likely the source of these tree topology differences.  Since in my
(limited) experience topological discrepancies are due to "treespace
searched" and not data, I don't think quartet puzzling methods would
produce a tree that had substantial internal support that is also
significantly different from one you would get using other "normal" ML
methods (using the same data set).


Byron J. Adams
Department of Plant Pathology
406 Plant Sciences Hall
P.O. Box 830722
Lincoln, NE 68583-0722
lab (402) 472 5598
fax (402) 472-2853

More information about the Mol-evol mailing list