On 27 Feb 2000, Sean Turner wrote:
> If you were to conduct a Kishino-Hasegawa test on the set of p-trees, I
> fully expect you would find the majority are significantly different from
> the best of the lot, which itself is liable to be significantly different
> from a fully optimized tree, yet all p-trees are given equal weight in the
> computation of the quartet puzzling values.
I think that depends entirely on the data set. Arndt von Haeseler has
done some (yet unpublished) investigations on the nature of the p-step
trees in order to be able to determine "online" how many p-steps
are actually necessary. One result is that in many cases the ML tree
*is* among the p-step trees - though sometimes only in a single copy.
Second, the frequency of occurence of a p-step tree depends on the data
set. If the sequences are very similar (low phylogenetic information)
then the tree frequencies decrease. However, there are many cases
where the set of p-step trees only contains very few different topologies.
So even though all p-step trees are given equal weight
in effect still a weighted consensus is computed: the individual
different p-step topologies are weighted by their respective
frequencies in the set of p-step trees.
I agree that there is potential in choosing a different consensus scheme,
or reoptimizing topology/likelihood.
Korbinian Strimmer http://users.ox.ac.uk/~strimmer
Dept. of Zoology, Univ. of Oxford, South Parks Road, Oxford OX1 3PS
+44 1865 271272 (phone), -49 (fax), korbinian.strimmer at zoo.ox.ac.uk