Hi all,

I've been recently looking at HyPhy documentation but I cannot find some of the models/options stated there. One of the options I've found in the documentation was the "Rate Heterogeinity" models (M0, M1 ... M5). Is there a possibility of setting these options through the GUI or do I need to go into coding?

Another question is concerned with the initial values hyphy assumes. Since there is the possibility of local maximums in the likelihood distribution, does hyphy makes any re-run with diferent starting vlaues?

At last, concerning LRT and the Akaike score, does HyPhy has any GUI implemented option to apply them?

Thanks in adv,

FG

Dear FG,

The documentation is woefully out of date, sadly. What exactly are you trying to do with the rate variation models?

Regarding initial values: some analyses allow one to specify them or pick them randomly, but most will pick some deterministic default values. The latter is the case with the GUI based analyses; but you can always tweak the values directly in the parameter table, and optimize starting from that point. If you are concerned about convergence, you can script up a follow-up run to 'nudge' estimated parameter values randomly, and re-optimize. I can help you with that if you had a specific analysis in mind.

Regarding LRT and AIC in the GUI, I would direct you to the HyPhy book chapter, which discusses how to set up nested models and compute LRT via the GUI (Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login).

Cheers,
Sergei

Hi all,

My problem is that I dont know, in the defined models, which  heterogeinity model is being used. The M0 must be used in the "Global" type models but the others..
And how can I set the M1, M2, M3 and M4 models?

Regarding the initial values I was thining of some kind of option that would allow for the automatic probe of several initial values of omega,kappa,etc.. in order to avoid local maximums.

thanks in adv,

FG

Dear FG,

I am still not sure which M0-M4 models you are referring to. I suspect you may be looking at the old documentation of models we decsribe in Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login. These are used by CodonRateVariation standard analyses and have more descriptive names. I never implemented these models in the GUI (and I should). Let me add them and get back to you in a bit.

You can script up the automatic sampler of initial values in the batch language (and I can show you how), but it's not very easy to do through the GUI, apart from entering the initial values in a parameter table and optimizing several times this way (you can save all parameter value estimates for each run there as well).

Generally speaking, if you suspect many local optima, a more general robust search approach (e.g. a genetic algorithm or simulated annealing) is probably called for. Unfortunately, these methods take a long time to run, and may be impractical for complex codon models, although they can be implemented in HyPhy with a bit of work. I'll be interested in seeing some examples of bad local optima traps if you have them.

Cheers,
Sergei