Problems with hitting distribution of real data

[miguelcama...@gmail.com]

Dear Dr Lopes,

I am using DNA sequences, 19 polymorphic loci, to estimate parameters of an IM model in two populations of a species which are isolated in two mountain tops. We have the hypothesis that the two populations split within the last 30ky and are nowadays exchanging genes with low frequency. Structure analysis clearly separates them as two populations and 2mod analysis support an IM scenario. I am having problems to set priors that actually lead to simulations that are close to my real data. We have been playing with different ranges in priors, tolerance in the rejection step and statistics of the .sst file. I further explored the statistics of the simulations with PCAs, but my real data is always very distant from the cloud of points. I am confident not to be breaking the assumption of the model of non-considered populations.

Do you think I might be misespecifying priors? Do you have any suggestions on how to proceed or explore what priors should I use?

Could the problem be related to other phenomena such as introgression from another species? (I highly doubt this is the case as the different alleles for the same locus usually diverge in very few sites, compared to others of close species)

I attach one of the set of priors I have used:

RUN INFORMATION

---------------------------

Number of geneological trees created: 100000

Populational tree topology:

- Single topology

Efective population size:

- Prior distribution (Ne1):    uniform(10,10000)

- Prior distribution (Ne2):    uniform(10,10000)

- Prior distribution (NeAnc1): uniform(10,100000)

Time events:

- Prior distribution (tev1):   uniform(0,100000)

Migration rates:

- Prior distribution (mig1):    uniform(0,0.005)

- Prior distribution (mig2):    uniform(0,0.005)

Mutation rates (Microsatelites):

- No mutation -

Mutation rates (Sequence data):

- HiperPrior distribution:     normal(4e-07,0,0,0)

Recombination rates (Microsatelites):

- No recombination -

Thanks, 

Miguel

[Joao Sollari Lopes]

Hi Miguel,

You should set some variation to the prior of the mutation rate
something like
normal(4e-07,0,1e-08,0)
or whatever value you think is best.
When you don't know much about the data, wider priors are best.

Hope this change will solve the problem.
Best,
Joao