Menu

convergence issues in BP&P3

BP&P Q&A
Anonymous
2015-03-13
2016-03-29
  • Anonymous

    Anonymous - 2015-03-13

    Hi,

    I'm trying to do a species delimitation analysis on 8 quiet recently
    diverged species with 4 loci. I have estimates of locus rates and a very
    unprecise root-age from a *BEAST analyses (the root was estimated to be 20
    My ± 15 My).

    I did a preliminary analysis with BP&P3 without delimiting species to
    estimate the theta and tau priors. Therefore I adjusted the tau prior
    according to the BEAST estimates:
    m = 20 My * mean rate of loci = 0.175,
    s = 15 My * mean rate of loci = 0.131,
    so I used G(1.78, 10.16) as prior distribution for tau0 and a diffuse prior
    G(1, 10) for theta.
    I fixed the locus rates and I also included heredity scalars since I
    included nuclear and mitochondrial loci.

    I wanted to use the estimates for theta and tau from this BP&P run in
    subsequent analyses with different guide trees and other modifications.
    However, no matter how long I ran the chains (50 000 or 500 000
    generations), I got very low ESS values <10 for the theta values of 2 or 3
    single prior species/populations and also low ESS values ~80 for tau at the
    deepest nodes. I checked the output in Tracer which showed heavily bimodal
    posterior distributions for the bad thetas.

    Is there an explanation for this behavior? Am I doing something wrong?

    Thank you very much for your help and time!

    Jonas

     
    • Ziheng Yang

      Ziheng Yang - 2015-11-13

      i am not sure, but one possibility is that beast defines parameters differently so that your priors may be off by a few orders of magnitude. perhaps you can look at the tutorial at the web site and run analysis A00 to get rough estimates of parameters.
      Also always start from the simplest model. turn off locus rates and heredity scalars. use those options only if it is necessary.
      ziheng

       
  • Anonymous

    Anonymous - 2016-03-13

    Is there any magic to get Tracer to read mcmc.txt files from A00 analysis? Tracer always tells me there is a formatting problem in Line so-and-so, which I can never find with text editors - all appears formed correctly,yet Windows versions of BPP and Tracer don't seem to cooperate. It would certainly make playing with parameters easier to consider...

    thanks

    Tom

     
  • Ziheng Yang

    Ziheng Yang - 2016-03-29

    the mcmc.txt file from the A00 analysis should be readable from tracer. i just tested and it did work on my laptop running win10. i have tracer 1.4. do you have a newer version?
    ziheng

     

Anonymous
Anonymous

Add attachments
Cancel





Want the latest updates on software, tech news, and AI?
Get latest updates about software, tech news, and AI from SourceForge directly in your inbox once a month.