MrBayes: Implementing the LG model

Question

[Adrian Pelin]

Hello,

When I run prottest on my amino acid alignment I am told LG+G to be the best model, which is common among amino acid alignments.

When I look at MrBayes documentation, I cannot find out how to specify the LG model. Does it need to be done somehow manually since it is not implemented into the program? How would I do that?

Adrian

Answer 1

[Adrian Pelin]

I got a reply from one of the developers if anyone wants an answer to this:

Hi Adrian,

The LG model is implemented in the developing version (http://sourceforge.net/p/mrbayes/code/HEAD/tree/trunk/). After downloading and compiling by yourself, just like setting the other AA models, use "aamodelpr=lg" to invoke the LG model.

/Chi

[Istvan Albert]

thanks for following up and posting the answer once your found out!

[Michael]

Hi Adrian,

Could you ever make this work? I tried with the devel build but it doesn't seem to work:

MrBayes v3.2.4-svn(r961) x64

prset aamodelpr=lg

No valid match for argument "lg"
   Invalid Aamodelpr argument
   Error when setting parameter "Aamodelpr" (2)

From the documentation help prset:

Aamodelpr      options:  Fixed/Mixed              default:    Fixed(Poisson)

and I think LG is a substitution matrix?

[Adrian Pelin]

See if this works, it's from my .nex file.

begin mrbayes;

prset aamodelpr=fixed(lg);
lset nst=6 rates=invgamma ngammacat=4;
mcmc ngen=2500000 samplefreq=250 printfreq=1000 nchains=8 temp=0.2 savebrlens=yes starttree=random;
set seed=21343;
sumt burnin=5000;
sump burnin=5000;
end;

[Michael]

thank you, yes the correct syntax is > prset aamodelpr=fixed(lg) not prset aamodelpr=lg

[Michael]

Thank you for your help,

tried this, the settings seem to work, but I have AA data so setting nst doesn't have an effect. My winner model from prottest is in fact LG+I+G so I set rates=gamma. No idea how to implement the 'I' (invariable sites) though. If you have any, I would appreciate the input.

Btw, if your model was LG+G shouldn't you set rates=gamma also?

[Adrian Pelin]

the nst=6 is redundant, I just kept it there as a template in case i decided to use nt. Let me share something cool:

I            : lset rates=propinv;
G            : lset rates=gamma;
F            :  prset statefreqpr=fixed(empirical);
I + G        : lset rates=invgamma;
G + F        : prset statefreqpr=fixed(empirical); lset rates=gamma;
I + F        : prset statefreqpr=fixed(empirical); lset rates=propinv;
I + G + F    : prset statefreqpr=fixed(empirical); lset rates=invgamma;

so as you can see, invgamma is LG + I + G when you set LG as your model.

[Michael]

these equations are indeed awesome, where did you get them from?

[Adrian Pelin]

From another post here on BioStars, but I don't remember where, I just wrote them in a file somewhere for future reference. I suppose we could email their list to validate this, I will give it a try and will report back.

Edit: Mr. Bayes Protein Models OR How to implement prottest output in mrbayes? here is where they are stated.

[Chirag Parsania]

Hi, Can I use FreeRate model as given here in IQTREE (http://www.iqtree.org/doc/Substitution-Models). I ran the IQTREE to generate the MaximumLikelihood tree. For the given dataset, IQTREE ran ModelFinder to choose best model to my data, which is LG+R6. Now I want to generate Bayesian tree using MrBayes with same model criterion. I found how to set LG in MrBayes. But what about remaining R6 which is FreeRate model.