Don’t you just hate when you run J model test (or similar software) to find the most parsimonious substitution model for a given set of sequences and the best model is something obscure and often not directly implementable in phylogenetics platforms like BEAST?
I stumbled across this excellent post by Justin Bagley that provides really useful information on how to put all sorts of substitution models in BEAST: http://www.justinbagley.org/1058/setting-dna-substitution-models-beast
This is definitely a valuable resource and makes incorporating J model test results much easier.