Hello

First many thanks for your software.
Just a couple of question:

Is there a maximum number of partitions that you can attribute to a dataset? I'm stuck with 5... Is there a way around that?

I'm not really familliar with those kind of analysis, and I'm extremely confused by the possible differences between some of the models tested in PAML, and the ones in hyphy. Is there major differences between the 2 programs for similar models? I guess that the "branch-site" model ("B") from PAML is implemented in hyphy, but where? Or do we have to define it?

Thanks for your time.

Dear joao,


wrote on Apr 7^th, 2005 at 9:31am:

Hello Is there a maximum number of partitions that you can attribute to a dataset? I'm stuck with 5... Is there a way around that?

There is no limit of the number of partitions attached to any given dataset. If you are working through the graphical interface, there is a limit that any given site can only belong to 4 partitions (i.e. partitions may overlap, but not more than 4 deep).

Are you having trouble defining more than 5 partitions? If so I'd be interested to know how...


wrote on Apr 7^th, 2005 at 9:31am:

I'm not really familliar with those kind of analysis, and I'm extremely confused by the possible differences between some of the models tested in PAML, and the ones in hyphy. Is there major differences between the 2 programs for similar models?

Take a look at Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login - major differences are discussed in the manuscript there (pgs 16-17). If you have questions about a specific model, I'll be happy to elaborate.


wrote on Apr 7^th, 2005 at 9:31am:

I guess that the "branch-site" model ("B") from PAML is implemented in hyphy, but where? Or do we have to define it? Thanks for your time.

HyPhy does not include predefined analyses to implement the branch-site models by Yang and Nielsen. It defines model a-la Yang 1998 (i.e. branch specific selection) - details in Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login and some others. It wouldn't be too difficult to write a 'branch-site' model in the HyPhy batch language though - I can help you get started if you want.

HTH,
Sergei

Further on this topic, what are the practical restrictions in terms of time and memory? How do these scale with nr of partitions, particularly when using codon models, and is there a rule of thumb recommendation for the maximum feasible number?

Konrad

Dear Konrad,


wrote on Oct 19^th, 2005 at 8:46am:

Further on this topic, what are the practical restrictions in terms of time and memory? How do these scale with nr of partitions, particularly when using codon models, and is there a rule of thumb recommendation for the maximum feasible number?

Each partition has its own tree with models/rate matrices/parameters (which you can constrain) etc. I have done some analyses with up to 10,000 (on two sequences) partitions in 1GB of RAM. Moreover, if you have access to an MPI cluster, you can use (an undocumented) MPIPARTITIONS command line flag to have HyPhy spread likelihood evaluations over multiple nodes (you need at least as many partitions as there are nodes).

Cheers,
Sergei