Hi,

I was wondering if anyone could help me create a slightly different version of CodonModelCompare.bf for HyPhy 2.0. As far as I understand it, the current CodonModelCompare.bf fits all 203 models mated with MG94, with a global dN/dS ratio on the tree.

What I'm trying to do is is fit all 203 nucleotide models mated with MG94_3x4, with local dN and dS parameters across the tree. Although this is computationally a bit expensive, the MPI version speeds things up enough to make it realistic for my dataset.

Apologies if this is a dumb question - I am almost entirely new to HyPhy.

Cheers,

Rob

Hi Sergei,

Thanks for the reply. I've tried it on a test dataset and it seems to work perfectly.

One further (unimportant) question - the "Model Options" prompt  still wants me to choose from  "Global" "Global w/rate var"  and "Non-syn variation only". As far as I understand I just have to choose "Global" (which now actually specifies a local model).

Just for completeness (and so I don't get confused when I try to use this in future), I tried to re-code this bit to skip that prompt and just go ahead and fit the local models (which presumably can't include rate-var. anyway).

I did this by deleteing this line:

#include "TemplateModels/modelParameters5.mdl";

and replacing it with

modelType = 0;

That seems to work - just wanted to check that it seems OK to you.

If that's OK, i'll post the bf to the appropriate list on the forum, in case anyone else would find it useful.

Thanks for your help,

Rob

Hi Sergei,

Thanks for the help. I have one observation which I don't understand.

If I use the new batch file, the output from the full GTR model is this:

(012345) Full Model ln-lik =  -67084.5. Parameter Count=271 AIC = 134711

As I understand it, this model should be MG94_3x4 mated with 012345 (it does at least have the correct number of 271 parameters). However, if I implement this model on the same data, with the same tree, using AnalyseCodonData.bf, I get a much better likelihood of -67002.5. I've had a look at both files, and can't see any difference in how they build the models. Also, I've tried running all the models in AnalyseCodonData.bf, and can't get a likelihood of -67084.5 with any of them.

This is not a particularly important problem for my purposes, but I'd be interested to know if there was something obvious which gives rise to this difference.

Either way, the new bf will be very useful.

Thanks for your help.

Rob