Setting up models such as HKY + I or GTR + I

[David J]

Could you please help. In beast version one I just clicked the invariant box.

In beast v2 there is gamma by default and then a rate for invariant. How do I set up a simple HKY + I model?

What do I set the below to for any model with a +I

Gamma category count
Proportion invariant (with estimate)

Additionally how do I find out the best suited invariant rate?

Thanks

[d1993]

Hi, 

I'm the new user of BEAST but I'll try to help. As I understand do you want to use the substitution model without gamma distribution? In my opinion, when gamma category count is "0" than you are not using gamma distribution rate. And information about invariant rate you should find in output file from the program to selection of best-fit substitution model for yours data, e.x. in jmodeltest2.  

[David J]

Hey d1993,

Thanks that is what I assumed. So you would use the invariant proportion given in J-model test then under the correct model. I assume you wouldnt click estimate then?

The part that is confusing is J-model test gives values of less than one for the gamma shape and the Gamma category count in Beauti is looking for whole numbers.....

[Andrew Rambaut]

Generally a model testing program is telling you which model fits best to the data not specifying what the parameter values should be. So you should probably let the alpha (gamma shape parameter) and pInv parameters just be sampled with the rest of your parameters.

The category count is not a parameter of the model but simply how many discrete categories of rate the model uses to approximate the continuous gamma distribution. The more categories you use the better the approximation is but the slower BEAST will take to run. The thing to realise is that we don’t need this to be a good approximation as we don’t believe it is really gamma distributed. Most people use a value of 4-8 for this. Probably the best thing is to use the same as J-Model test.

Andrew

--
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.

[David J]

Thanks Andrew.

So if J model test indicates a model with + I do I just set the invariant proportion to 0.1 and estimate? 

In beast V1 I would just drop down the tab to the invariant option... It’s a bit more overwhelming in V2. If you me a good link for understanding the set up of the model I would appreciate it. 

Thanks

Kind regards,

David 

[Andrés Parada]

Hi,

In BEAST2 you just click or select estimate 'Proportion of Invariant' in the Site Model tab. Then you click or drop down the "pencil" next to the 'red button' and put a very small number like I did in the example below: 1 e-7 (this is just a value to start the estimation, I remember this advice from an old answer)

Cheers

Postdoctoral Associate

Instituto de Cs Ambientales y Evolutivas

Universidad Austral de Chile

Campus Isla Teja s/n

Valdivia, Chile

[David J]

Thanks Andrés that’s is an awesome help. 

Kind regards,

David Juszkiewicz

On 7 Feb 2019, at 00:16, Andrés Parada <andre...@gmail.com> wrote:

Hi,

In BEAST2 you just click or select estimate 'Proportion of Invariant' in the Site Model tab. Then you click or drop down the "pencil" next to the 'red button' and put a very small number like I did in the example below: 1 e-7 (this is just a value to start the estimation, I remember this advice from an old answer)

Cheers

<image.png>