Menu ▾ ▴

Model selection

Users
TreeFam
2013-10-01
2014-08-25

  • TreeFam

    TreeFam - 2013-10-01

    Hi guys,
    we would like to test CodonPhyML on a large dataset of ~16,000 gene families with ~100 members each.

    In the paper you mention that you are planning on implementing automated model selection. Is that already implemented?
    If not, what is the best way to select a suitable model for a given gene family?

    Thanks a lot for your help!

    Cheers,
    Fabian

    --
    Dr. Fabian Schreiber
    TreeFam Project Leader

    The Wellcome Trust Sanger Institute /
    European Bioinformatics Institute

    e: fs9@sanger.ac.uk
    e: fs@ebi.ac.uk
    t: 01223 494726
    w: www.treefam.org/
    --

     

    • Marcelo S Zanetti

      Marcelo S Zanetti - 2013-10-01

      Hi Fabian,
      Thank you for your question. I forwarded it to my colleagues. Will come back to you asap.
      If you have further inquiries do not hesitate to contact us.
      Best

       

  • Stefan

    Stefan - 2013-10-01

    Hi Fabian

    As for the first question - no, model selection is not (yet) implemented. Still on the todo list. When we have to do this, we usually use the following approach:

    • if the models are nested, you can use a hierarchical LRT setup (Posada and Crandal, 1998), a dynamical LRT setup (Posada and Crandal, 2001) or backward elimination (Bao et al, 2007)
    • if the models are not nested, AICc or BIC

    You could also apply several methods and come up with some methods to compute a consensus.

    I hope this helps. If not (maybe because we misunderstood your question), please tell us.

    best
    stefan

     

  • Stefan

    Stefan - 2013-10-01

    EDIT:

    Ouch, I forgot: if you want to compare nucleotide/aa/codon models, Seo/Kishino (2009, sysbio) could be of interest. Furthermore, we suggest to do all this analyses on a fixed topology (branch lengths can be re-estimated, of course).

    best
    stefan

     

  • Anonymous

    Anonymous - 2014-08-21

    I am also interessed in a tool that would select the best codon model to use for my data set. You wrote soon a year ago that that was on the todo list. Have you made any progress on that? If you do note have such a tool yet, have you noticed any model that seems to outperfor the others that I could start with?
    Greta

     

    • Marcelo S Zanetti

      Marcelo S Zanetti - 2014-08-25

      Hi Greta,

      Thank you again!

      The model selection is still in the TODO list :-), sorry for that.

      To help you choosing a model I will need to know which kind of data you plan to work on. Please, let me know.

      And again, if you need anything else write us.

      Best

      Marcelo

       

Anonymous
Anonymous

Add attachments
Cancel