question about MrBayes software: evolutionary models

Question

[gobiidae.marin]

I'm a PhD student and read your tutorial about MrBayes softwar. for my thesis, I working with MrBayes software to analyzing data. your help and guidance are valuable for me. I have a question about substitution model for mrbayes software. I use jModeltest to determine the best model for my sequences. and the TPM1uf + I + G is the model I got form jModeltes (BIC).
MY question is: How can I apply the model which determine by jModeltest in MrBayes software.

for example : the model TPM1uf + I + G and parameters: nucleotide frequencies A: 0.27, C: 0.19, G: 0.17, T: 0.37; substitution rates RAC: 1, RAG: 13.33, RAT: 4.63, RCG: 4.63, RCT: 13.33, RGT: 1; gamma distribution shape = 0.48; and proportion of invariable sites = 0.38 .

Is it mean that I apply the substitution rates through Ravmatpr=fixed (1,13.33,4.63,4.63,13.3,1) or Ravmatpr is different parameter from substitution rate?
and is it necessary to apply nucleotide frequencies by statefreqpr=fixed(0.27, 0.19, 0.17,37) also for gamma distribution shape and proportion site ? also, If I used nst=mixed to estimate substitution model by Mrbayes, it is still necessary using Jmodel test or mrmodeltest?

thanks in advanced

Answer 1

[17th Level Botanist]

Old question, but deserves an answer in case others search it.

Here is a list of models that can be applied in MrBayes. There are two main methods to pick one of these models from among your results. You can pick the highest scoring that you can actually set in MrBayes or you can use the model that is most similar to your highest scoring model.

However, arguably, you do not necessarily need to do model selection for MrBayes. You can just use GTR+G, as all other models are nested within it and Bayesian analyses do not suffer the over-parameterization problems suffered by ML. The only time not to use GTR+G s is you do not have enough examples of a particular kind of substitution in your data (e.g., if you have none or only a few of C->A). With respect to when not to set the GTR + G model, see the MrBayes wiki, which states:

Before using the GTR model for some of your data, we recommend that you make sure there are at least a few possible substitutions of each type. For instance, if there is not a single GT substitution in your data, it will be difficult to estimate the GT substitution rate. In such cases, you should consider either the HKY model (nst=2) or the F81 model (nst=1) instead of the GTR model.

A final note that the +I parameter is not needed. As +G can be 0, there is no need for +I, and +I can sometimes actually interfere with estimation of +G.