Nucleotide substitution model and neighbor-joining tree

Good morning!

I would like to build neighbor-joining tree, but can’t decide what nucleotide substitution model should be used.

As I understand, fitting nucleotide substitution models in MEGA is option for distance-based methods (this post). But does it really matter to character-based like NJ? If I choose NJ-method in building phylogeny section, different replacement models are offered. And resulting trees have different branch supports depending on the model! So, it is important.

Having made sure of this, I choose Models > Find best DNA/Protein models (ML) (yes, ML, I would like to use NJ, but futher there is NJ option? ->), then Tree to use – Automatic (NJ tree) and obtain table with GTR, GTR+I, GTR+G, GTR+G+I, K2, K2+G, K2+I, K2+G+I, JC, JC+I, JC+G, JC+G+I, T92, T92+G, T92+I, T92+G+I, HKY, TN93+I, TN93+G, TN93, HKY+I, HKY+G, TN93+G+I, HKY+G+I, whereas in building phylogeny window (NJ) there are such options: No. of differences, P-distance, Jukes-Cantor model (JC), Kimura 2-parameter model (K2), Tajama-Nei model (where is estimation?), Tamura 3-parameter model (T92), Tamura-Nei model (where is estimation?), Maximum Composite Likelihood (where is estimation?), LogDet (Tamura-Kumar) (where is estimation?).

I have also tried modeltest (phangorn), but can’t find estimation for Tajama-Nei model, Tamura 3-parameter model, Maximum Composite Likelihood, LogDet (Tamura-Kumar), and can’t determine Tamura-Nei = TrN or TrNe in it’s output (separate post).

Could you please help me to choose the nucleotide substitution model for NJ?

Thank you in advance!

Best regards,
Poecile

Read more here: Source link