does the best substitution model for a phylogeny tree differ between one method to the other ?

Question

[Michel Edwar]

I am using MrBayes, PhyML and phangorn to build the trees for a certain data set. I have used ProtTest and it identified LG+I+G+F to be the best substitution model for the maximum likelihood method (PhML). My question is, can I use the same substitution model in Mr Bayes (Bayesian analysis) and phangorn (nearest neighbor) , or do I need a different test for each method?

Answer 1

[h.mon]

You may use the same model for MrBayes (if possible to specify this model in MrBayes), or use MrAIC to find the best model.

Nearest-neighbour is a method to perform tree rearrangement to search the parameter space, no relation to the evolutionary model you choose.

[Michel Edwar]

but in MAFFT server, you can identify one of three substitution models for neighbor joining method.

[Substitution model (valid when NJ is selected): JTT
WAG Alpha
Poisson]

[h.mon]

Not all programs implement all models. For example, I do not know if the latest MrBayes implement all models tested with ModelTest, I know an earlier version (I think 3.1 but may be wrong) implemented only a subset of the models found in ModelTest.

Anyway, if you are doing ML and bayesian phylogenetic inference, there is no need for NJ, which in general is less accurate.