In article <8hmopn$l61$1 at mercury.hgmp.mrc.ac.uk>, Dmitri Sherbakov
<sherb at srv01.lin.irk.ru> wrote:
> I came across a dataset of 18S rRNA sequences of algae, where phylogenetic
> analysis suggests the following:
>> A->B->C->D (1)
>> instead of "normal"
>> A B C D (2)
> ! ! ! !
> +-+-+ ! !
> ! ! !
> +--+--+ !
> ! !
>> due to zero or almost zero lengths of some branches.
> Since the case (1) may be describe realistic evolutionary events, the
> questions are:
>> Is it possible to built a test distinguishing between (1) and (2)?
If you are saying that tree 2 is your null hypothesis which you wish
to try and reject in favour of tree 1, then this is simply a test of
rate constancy (molecular clock) and you can do this using a likelihood
ratio test. Assuming tree 1 is your maximum likelihood tree, estimate
the likelihood of tree 2 (using the molecular clock option of PAUP,
PAML or using DNAMLK in PHYLIP). Take twice the difference in the log
likelihoods between the two models. If this is greater than the
Chi squared critical value (at the alpha you are interested in) with
(in this case) 2 degrees of freedom (more generally n-2 d.f. for
n taxon trees).