I am doing phylogenetic analysis of a highly conserved protein's AA sequences over a wide range of Eukaryote taxa using MrBayes' mcmc method. MrBayes, when run in mixed mode, consistently chooses the rtREV model, while ML methods (prottest3 and iqtree choose LG(+G) or WAG+(G) as best scoring model). The rtREV phylogenies look generally "better" in both Bayesian and ML, they group clades more in accordance with expectations and manage to reconstruct more - and more sensible, bifurcations - whereas the MrBayes phylogenies are generally "better" than ML. Choice of model was independent of the multiple sequence alignment software, me editing alignments, or other parameters (+I +F +G).
rtREV was initially developed for retroviral or reverse transcribed sequences, but my sequences are not known to be retroviral. My question is simply, does that choice "mean" anything, e.g.
- Are my sequences simply very diverse, like in retroviral sequences?
- Or should I look into something else, like horizontal gene transfer via retro-viruses or retro-transposons
- Something else? Or nothing?