SNAPP Theta Prior for BFD

[Kenji Yip Tong]

Hello Everyone,

I am currently trying to perform a species delimitation study using Bayes Factor Delimitation (BFD). In such a study, it is essential to provide rational prior values. 

For the prior on the coalescent rate Theta, It seems I can only set one distribution. 

Given that there are multiple species/populations in our dataset, each with their own specific Theta, which value should I use? 

I am thinking of taking the average Theta of all my populations, and have the prior distribution reflect the distribution of Theta values.

I haven't seen any mention of this issue anywhere. Am I missing something? 

Thank you very much in advance. 

Kenji. 

[Remco Bouckaert]

Hi Kenji,

The prior distributions supported in BEAUti are indeed limited. One of the reasons is that internal node branches can move around during the MCMC making it cumbersome to specify individual branch rate priors. I suppose that if the theta values you know a priori for the different species are not too far apart (i.e. not many orders of magnitude) taking the mean of them makes sense for the theta prior as long as the variance of the theta prior is at least large enough for the theta prior to cover all values. 

Regardless, you can check the posterior under different priors to see how sensitive the posterior is to your theta prior — perhaps your data has sufficient large signal that the prior does not matter very much.

If you really want to specify priors for individual species, this can be done by editing the XML and for example using the ExcludablePrior from the BEASTLabs package (which is already installed if you use SNAPP). Just add the XML fragments similar to the ones below inside the distribution element with id=“prior”.

This specifies a gamma(1,2) prior on the first dimension of the theta.s:dna parameter:

            <distribution id="Theta1_prior" x="@theta.s:dna" spec="beast.math.distributions.ExcludablePrior" xInclude="1 0 0 0">

                <distr spec="beast.math.distributions.Gamma" alpha="1" beta="2"/>

            </distribution>

The text that is bolded is what you want to adept for your own prior. This specifies a gamma(2,2) prior on the second dimension of the theta.s:dna parameter:

            <distribution id="Theta2_prior" x="@theta.s:dna" spec="beast.math.distributions.ExcludablePrior" xInclude="0 1 0 0">

                <distr spec="beast.math.distributions.Gamma" alpha="2" beta="2"/>

            </distribution>

And this specifies a gamma(3,2) prior on the third as well as the fourth dimension of the theta.s:dna parameter:

            <distribution id="Theta34_prior" x="@theta.s:dna" spec="beast.math.distributions.ExcludablePrior" xInclude="0 0 1 1">

                <distr spec="beast.math.distributions.Gamma" alpha="3" beta="2"/>

            </distribution>

Hope this helps,

Remco

--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beast-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/5d102ec7-acec-460b-b2b6-b8c74e1e3431n%40googlegroups.com.