Hi. I don't have an answer to your question, but a tip. If you use partitionfinder instead of jmodeltest, then you can restrict the outcome models to those implemented in the phylogenetic analysis programs. E.g. you can set in partitionfinder 'give only those models I can use in MrBayes', and it will. You may not use the same model as jmodeltest suggests, but I guess you will have trouble running the less common models anyway.

Like Rebecca, I have never heard of this model, but find it strange that jmodeltest would implement a model that is not used for inference in standard programs. Anyway, and along the same lines as Rebecca, you can simply move down the list of recommended models in relation to their likelihood score until you come to a model that can be used. Or you can implement your own model (012030) yourself in most programs. Also from my experience, the shorter the alignment the more likely you are to receive a strange evolutionary model result. An example of this is with nucleotide sequences from eukaryotes, if you are not getting the GTR as the most optimal model then something strange is happening. Also what algorithm are you getting this result from? AiC, BiC? rates test? hierarchal? this is also important to consider.

regards

Thanks Rebecca and Russel. My sequences are around 1.4kb long. The jodeltest2 proposed best model 012030+G+F was using AIC.

Its strange and disappointing as well that there is no strong and clear documentation with jmodeltest2 that clarifies all models details. There are a number of such models described as code like 012000, 001020, 001234, 012340, 012341, 012342, 012343, 012344 etc .... but specially for 012030 I could not find any information what is this.

Anyway, fortunately my second best model is TrN+G and I will go for that as you suggested. Cheers.

Too late an answer, but you could use PhyML that allows you to input the custom model of interest.