Hi,

I was surprised to find that the MG model (modelType = 1 in NielsenYang.bf) runs more than twice as fast as the GY model (modelType = 3) on an example analysis (1 hour 40 mins vs 4 hours). This seems strange because, if I understand it correctly, the GY model is a special case of the MG model, with the synonymous rate constrained to be constant, and should therefore have an easier optimisation task. Does this sound believable?

thanks,
Konrad

Dear Konrad,

In the context of NielsenYang.bf, all models have constant dS, and the differences between MG94 (classic MG94) and GY94 is that (1). GY corrects for transition/transversion bias (MG94 does not, but MG94custom x 010010 does, and it can correct for other nucleotide biases with the appropriate options) and (2). GY considers rates to be proportional to the frequency of the target codon, whilst MG assumes that they are proportional to the frequencey of the target nucleotide.

I am not sure why the run times are that different - it may have to do with guessing the starting point, and perhaps affecting the rate of convergence in the omega distribution parameters.

If you want to model dS and dN variation jointly, check out CodonSelectionAnalyses:dNdSRateAnalysis.bf

HTH,
Sergei