Low prior and posterior ESS values

[Fran3]

Hi all,

I hope this message finds you well!

I'm writing because I'm ahving some problems with my data. I have historical and modern complete mtdna samples and I performed the analysis on BEAST in this way:

- I put tip dates for all my samples as years before present

- I created two taxon sets for which I forced monophily and I put divergence time for each taxon in years before present (with a normal distribution and a st.deviation both in million years)

- I used a GTR+G+I (empirical) substitution model

- strict clock

- clock rate: Uniform [0, 1E100], initial = 1

- treeModel.rootHeight: using tree prior in [119, infinite]

- constant.popSize: I tried both Uniform, leaving default values, and also Lognormal [1,10], initial=1

Chain length 50.000.000, log every 1000

I always get low ESS values for prior and posterior. What am I doing wrong?

Thank you,

Fran

[Jordan Douglas]

Hi Fran,

The problem may be the clock rate prior being improper. The tree height and clock prior are usually strongly (negatively) correlated and therefore this causes mixing issues. I suggest you add an informed prior for the clock rate (ie. how many substitutions per site per million years do you expect).

Remco Bouckaert is in the process of writing help-me-choose guides for beast2. There is a good guide on selecting clock rate prior here

https://beast2-dev.github.io/hmc/hmc/Priors/ClockPrior/

Hope this helps

Jordan

[Fran3]

Thank you very much, I will give it a try!