Measure For The Separation Of Phenotypic Groups On An Unrooted Phylogenetic Tree

Question

[Will]

I'm trying to find a way to evaluate a phylogenetic tree for its ability to 'separate' a set of sequences into different branches of an unrooted tree. Each of my sequences has one of two phenotypes which was determined by a functional assay. In an ideal case they would be completely separated in an unrooted tree but in reality a branch that is primarily composed of phenotype-A will be contaminated with sequences with phenotype-B.

My current method for dealing with this is to calculate the distance between each sequence and a reference sequence and then looking for a difference in the average distance between the reference and phenotype-A sequences and phenotype-B sequences. However, this relies on the assumption that the evolutionary distance between phenotype-A and the reference are different between phenotype-B and the reference.

I'm hoping to find something that doesn't rely on a comparison with a reference but solely with the proportion of sequences from each phenotype in each branch of the tree.

Any ideas or references that someone can point me towards?

Answer 1

[Agapow]

In the past when I've had to do such a thing, I've measured a clustering or imbalance metric across the phenotypes in question and then compared that to trees in which the states were repeatedly randomized. Thus you get some sort of confidence for the separation of the two states against a case where there's no separation. The software BATS(done by Joe Something-or-other from Oxford in Andrew Rambauts group) will do this for you.

[Will]

http://evolve.zoo.ox.ac.uk/evolve/BaTS.html ... for those looking for it later.