problem with tree root

[Glémin Sylvain]

Hello,

I've recently started to use testNH with codon models.

As in examples, I first run bppml with a fixed YN98 model to get model parameters for mapping and also use the output tree (.dnd) as the input tree for mapNH.

I gave a rooted tree but the root of the output tree is different from the input tree.

How can I keep the same rooted tree?

Thank you

All the best,

Sylvain

[Julien Yann Dutheil]

Hi Sylvain,

May I see your tree (you can send it outside this forum) ? I'm wondering, as normally codon models are stationary and fitted on a non-rooted tree, if you would be using a non-stationary codon model (for instance with heterogeneity in GC and not in Omega) ? As far as I remeber, neither bppml nor mapnh reroot the input tree...

Best,

Julien.

--
 
---
You received this message because you are subscribed to the Google Groups "testnh-help-forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to testnh-help-fo...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

--
Julien Y. Dutheil, Ph-D
0 (+49) 6421 178 986

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

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

[Julien Yann Dutheil]

Forgot to ask: which version of bppml / testNH are you usIng ?

Cheers,

J.

[Bastien Boussau]

Hi Sylvain,

Maybe you ran testNH this way already, but to keep the root with a reversible model (as the YN98 model), you need to make as if your model was branch-heterogeneous. More clearly, you need to specify one model per branch, one set of equilibrium frequencies at the root, and then link the parameters together so that both the root and all the branches have the same stationary frequencies. This way bppml will keep the root, as will the subsequent programs in the pipeline.
I hope this helps!

Best,

Bastien.

[Glémin Sylvain]

[Glémin Sylvain]

Thanks Julien and Bastien for your answers.

I think it's OK with Bastien's solution.

I have an additional question because maybe I didn't understand well how to use the program.

I simply want to group branches according to similar omega and the initial mapping is done with a homogenous model.

However, I though that for the model choice procedure it was important to have the correct rooted tree. Is it correct?

To answer to Julien:

My bppml version is 1.5.0 and the testNH version is the 0.1.0

All the best,

Sylvain

On Thursday, August 29, 2013 9:54:07 PM UTC+2, Glémin Sylvain wrote:

[Bastien Boussau]

I think you should get extremely similar results with a rooted or with an unrooted tree. If you keep equilibrium frequencies constant over the branches of your tree and at the root, then I think your model should be reversible, in which case it should be insensitive to where the root is placed. If however you have e.g. different equilibrium GC contents in different parts of the tree, then it matters where the root is.

Finally, I would say that if you don't have varying equilibrium frequencies in your tree, you're probably safer using an unrooted tree. Otherwise you're asking the program to estimate parameters (position of the root between its two children) for which there is no information, because the model is insensitive to the position of the root.

Bastien.

[Glémin Sylvain]

OK

Many thanks

Sylvain

On Thursday, August 29, 2013 9:54:07 PM UTC+2, Glémin Sylvain wrote: