Mary K. Kuhner
mkkuhner at phylo.genetics.washington.edu
Mon Aug 28 08:54:22 EST 1995
Carla Ann Hass <CAH19 at psuvm.psu.edu> writes:
[citing someone whose attribution is gone]
>>Neighbor Joining is NOT a good tree building method. If you change the
>>order of input of taxa, you change the tree.
> What is your reference for this statement? I have "built" lots of neighbor-
>joining tress using both MEGA and other programs and have never found this to
Having recently written a neighbor-joining program (for PHYLIP) I
can say with fair confidence that the only effect input order has is
that it is used as a tiebreaker. For example, if several sequences are
identical they are represented as several successive bifurcations rather
than as a multifurcation (though we hope to change this in
a subsequent release as it can be misleading). The order of these
bifurcations is based solely on input order since the data are
Neighbor joining constructs a tree; it does not search among possible
trees. This actually makes it relatively independent on input order.
Programs which search among trees, and don't consider all of
them, generally have much more dependence on input order of species.
This is true of both the FITCH distance program and the DNAPARS
parsimony program in PHYLIP; it is not a specific problem with distance
methods. Any search that does not check all trees will miss some.
If one is concerned about this, one can run the program several times
with different input order (or using the Jumble option in PHYLIP) and
see if the result changes. It is also possible to use a program
which does a complete search and is guarenteed to find the optimal tree,
if the data set is small enough. PAUP does this well for parsimony
(much faster than PHYLIP in this regard).
Mary Kuhner mkkuhner at genetics.washington.edu
University of Washington
More information about the Mol-evol