Hi Sergei,
    We are finding the best nucleotide substitution model for a number of alignments using the best nucleotide substitution analysis (with gamma+inv rates) in HYPHY.  From this analysis, we get an estimate of alpha for each alignment.  However, the alpha estimate obtained using the best nucleotide substitution model does not always agree with the alpha obtained using the HYPHY GUI, where we input the chosen nucleotide model and use the built in likelihood method to estimate the rate parameters (including alpha).  Can you help? I have listed the steps involved for each method and can email a problem alignment if you'd like.
   
1 - Standard Analysis: NucModelCompare:-Global w/variation - Once - Gamma+Inv- 4 rate classes - significance of .0002;

2 - Hyphy GUI: created a new model (011010) from the NucModelCompare results with het. rates; applied this model to the whole alignment (build likelihood, optimize) and got an estimate of alpha.

The problem is that the two estimates of alpha do not agree. For one alignment, method 1 yielded an alpha of 2.15, while method 2 yieleded an alpha of 1.41.

Thanks very much,
Jennifer Knies