distance based DNA trees

[Don.Colless at CSIRO.AU]

Rob Smissen wrote:

------------------------------------------------------------------------

I published a nucleotide substitution distance tree last year, I may do
it again. I very seldom find neighbour-joining trees built using an
appropriate substitution model differ significantly from parsimony
trees. Ditto for least squares or minimum evolution trees. NJ has the
advantage of speed. Distance trees seem to be rather more realistic in
estimating internal branch lengths for ancient radiations as a bonus.
Where they do differ, there is usually a lack of clear signal in the
data and neither is at all reliable. Where different tree building
methods do give different results there is usually a specific
explanation - that is the disagreement reveals or at least illustrates
important properties of the data. Sometimes it will then be possible to
argue why one method is better in the specific case - I don't think you
can do that generally and absolutely.
-------------------------------------------------------------------------

How nice to find someone actually applauding the Neighbour Joining method. I've had the same experience: laboriously running PAUP under various options and finally getting practically the same result (or one that looks as good) - in a matter of seconds - with my NJ programme. I have it, by the way,  packaged with my UPGMA etc., a most enjoyable heresy.

Don Colless,
Div of Entomology, CSIRO,
GPO Box 1700,
Canberra. 2601.
Email: don.colless at csiro.au
Tuz li munz est miens envirun