Using bModelTest Results in BEAUti

[Michael Bose]

I am new to using bModelTest and have never had to use nonstandard substitution models in BEAUti.  If for example, bModelTest recommends model 123323 how do I interpret that into input for BEAUti? Or is that not a possibility and I will need to modify the BEAST 2 xml instead?  Thanks.

[Remco Bouckaert]

Hi Michael,

The bModelTest site model (from the bModelTest package in BEAST) allows averaging over substitution models — so, it won’t recommend model 123323, but possibly sample that model more often than others if that model is well supported by the data.

If you use jModelTest — a program separate from BEAST — it can recommend 123323 as the best fitting model. You can create this model by selecting GTR as substitution model and link the appropriate parameters (after you select the menu Mode/Allow parameter linking in BEAUti 2).

Cheers,

Remco

On 22/09/2016, at 9:22 AM, Michael Bose <mikeb...@gmail.com> wrote:

I am new to using bModelTest and have never had to use nonstandard substitution models in BEAUti.  If for example, bModelTest recommends model 123323 how do I interpret that into input for BEAUti? Or is that not a possibility and I will need to modify the BEAST 2 xml instead?  Thanks.

--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beast-users...@googlegroups.com.
To post to this group, send email to beast...@googlegroups.com.
Visit this group at https://groups.google.com/group/beast-users.
For more options, visit https://groups.google.com/d/optout.

[Michael Bose]

Hi Remco,

Thanks for the response.  Perhaps recommend was the wrong word to use.  If the results of bModelTest showed that model 123323 had 77% posterior support that would suggest that model 123323 fits the data best and is the one that should be used for the BEAST analysis, right?

Thanks,

Mike

[Remco Bouckaert]

Hi Michael,

With bModelTest as site model, you can do the analysis without committing to a specific substitution model. Uncertainty about the substitution model will be averaged out through MCMC. So, there is no need to fix the substitution model, though when there is such large support for a single model it probably won’t matter much if you do.

Cheers,

Remco

[Michael Bose]

Oh, I didn't understand that it could be used that way.  Based on its name and what I had read about it I assumed it was a tool like jModelTest for selecting the best fit model for the data.  Not having to select a specific substitution model at a all is definitely easier.  Thanks for the explanation.

Mike