Mark Siddall wrote:
>Neighbor Joining is NOT a good tree building method. If you change the order of
>input of taxa, you change the tree. Ummm... sound like a problem?
>If you replicate a taxon you change the tree elsewhere.
>There are manmy other problems.
is that the best criticism of NJ you can think of?
Are you seriously suggesting that this never happens with parsimony???
If you bootstrap your NJ tree (as you should do anyway) and JUMBLE the input
order of the sequences in each replicate (e.g. the J option in PHYLIP), the
problem goes away (for both parsimony and NJ). If you have some sequences
that are more or less identical, then the branching order between them
is arbitrary and is indeed an arbitrary function of input order. However,
the internal branch lengths connecting them will be zero. This is not
exactly a problem (providing you look at the branch lengths; some folks
do not seem to think that is useful though; I wonder why??).
Similarly, if you have some equidistant sequences, you can get an arbitrary
branching order (not always) but again, the internal branches will be close
to zero in length i.e. will show up as a multifurcation; the real answer.
- can you show me a REAL data set with REAL sequences where this
is a real problem for NJ and not parsimony i.e. where bootstrapping or
an examination of branch lengths will not tell you what is going on?.
The way you state it above, you imply that the branching order ALWAYS
varies depending on the input order ..... are you sure you are not
guilty of misinformation yourself??