very large omega values

[Tristan Lefebure]

Hi there,

On a big phylogeny with some short branches, testNH is getting me some weird omega values, the worst being 999. If one look at the relationship between omega values and branch length, it suggests that for some very small branches, probably with a very number of substitution, the inferred dN/dS is not reliable. What do you think? I attach a plot describing the problem. Probably not an elegant solution, but I was thinking of bootstrapping the whole process to see how robust is the clustering.

Thanks!

--
Tristan

[Julien Yann Dutheil]

Hi Tristan,

This is a very interesting point, indeed, very short branches do not
allow to estimate proper omega values, and are a factor of
overparameterization. In principle, PartNH should never favor a model
with such omegas associated to small branches. In practice however, it
all depends on the clustering, and it can happen that in some cases,
short branches are wrongly clustered. In mapnh, there are options to
say that short branches should be automatically clustered with their
parents. You can either say say the total number of events should be
higher than a certain threshold, or that each type of mapped event
should be documented by at least a certain number of observation. I
think this is in the documentation of mapnh. I hope this will solve
your problem.

All the best,

Julien.

> --
>
>
>



--
Julien Y. Dutheil, Ph-D
0 (+49) 6421 178 530

§ Max Planck Institute for Terrestrial Microbiology
Department of Organismic Interactions
Marburg -- GERMANY

§ Intitute of Evolutionary Sciences - Montpellier
University of Montpellier 2 -- FRANCE

[Giap Nguyen]

Good afternoon                                                                                                                                                  
Would you please give a command to tell mapnh to 'automatically cluster short branches with their parents'?

Thank you very much!

Best regards

Giap Nguyen

2012/8/30 Julien Yann Dutheil <jy.du...@gmail.com>

--

[Tristan Lefebure]

I think Julien is referring to the following command:

test.branch.auto_cluster = Global(threshold=1)

This will cluster any branches with 0 count with their parents. One has then to choose a threshold....

--
tristan

[Julien Yann Dutheil]

Hi,

Indeed. One can also use Marginal(threshold=1), to ensure all
categories are represented.
In practice, Marginal(threshold=1) and Global(threshold=3) are
reasonable choices (to my experience).

Best,

Julien.