Theta Priors for SSR data

[Darius Danusevicius]

Dear all,

I use MIGRATE to estimate effective pop size via theta with a data set on 11 SSR loci in several species of wild animals and trees.

The suspected problem is the range of the priors for theta. The defauts are 0 to 0,1 with mutipliction step of 0.01. I used the Brownian model (to save time) and with the default settings for the theta priors, I get suspiciously uniform theta of ca. 0.0980 pines, limes, alders even wolf.  The loci are polymorphic highly 11 to 18 loci. Population size ca. 100 indv.  MIGRATE also returns the warning of too low upper bound for the Theta priors. I have experimented with increasing the upper prior bound for Theta under a multiplication distribution to 1, 2, 3 4, 5, 6, 7, 8, etc. The problem is that the theta is changing each time I change the upper bound, which makes the result unreliable. Can you help with that.

Another problem that we tried to run the long microsatellite model. However, MIGRATE stops after calculating the allele frequencies. Why? Data is OK as it well runs under the Brownian model. May pop size too large 150 trees and 3 pops?

 

Darius

[Peter Beerli]

Darius

the easiest way to give guidance on this is by sending me (bee...@fsu.edu) the infile, the parmfile, and an output from one of your runs. I do not understand what you mean with "multiplication distribution” the uniform prior for theta is specified by low bound, upper bound , and stepsize, the stepsize

is to limit big jumps from one parameter change to the next in the MCMC, the smaller the stepsize is the smaller the changes, I usually use about 1/10 of the range.

Peter

--
You received this message because you are subscribed to the Google Groups "migrate-support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to migrate-suppo...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/migrate-support/938a4539-f53f-476e-827c-7d8c4dcd4ee9n%40googlegroups.com.

[Darius Danusevicius]

Dear Peter,

Thanks a lot for your response.

Multiplication distribution refers to the Theta prior distribution, when  I set the upper bound for the Theta prior distribution (Search menu) then it asks to choose between the Multiplication and Exponential distributions. 

Ima attaching the infile with one pop and the parameter file (and the output files). This parameter file is set for the uniform Theta prior distribution (i.e., as I understood when the Theta prior distribution setting is to uniform then no upper limit for the Theta prior is applied, however, as I leaned later that the upper limit is set to 0.1). 

So in the later runes I simply varied the upper limit for the Theta prior distribution from 1 to 10 by 1. This resulted in a variable Theta, depending on my settings of the the upper limit (not attached). Then the question is what is the reasonable setting o the upper limit of the Theta prior. Alternatively, may it better NOT to use the "Brownian motion" data type option for the SSRs but relay on the longer alternative called "microsatellite model"? However, again thus longer alterative may have the same problem of what is appropriate setting of the upper bound of the Theta prior? (and be tricky to optimize because of long running time)...

Darius

[Darius Danusevicius]

Here are 2 more runs with upper bound for Theta priors of 4 and 20

[Peter Beerli]

Dear Darius,

please download the current version of migrate — it is 5.0.4 from https://popgen.sc.fsu.edu or https://peterbeerli.com/migrate-html5/index.html

 (there is also an older version 3.7 but I am phasing that one out),

you ran your examples on version 3.6.

I have in the meantime, abandoned the multiplier prior because it did not work well. The uniform prior can be set between any min and max but need 

always specify a maximum because otherwise, the prior becomes improper.

For the SSR I would always use the Brownian model because it is fast and delivers similar result as the explicit stepwise model. the stepwise model is horribly slow in comparison; upon a quick test, I seem to have broken the single-step model in the newer version of migrate and will need to fix it. I suggest using the Brownian model — both models have nothing to do with the bounds for the Theta prior.

You may also want to look at a tutorial on how to use migrate, I also added a PDF of short note of the brownian motion vs single step model.

Peter