Extended Bayesian Skyline Plot trace jumping around after indicator fixes to 1

[Zach N]

Hello all!

I am trying to perform an Extended Bayesian Skyline analysis on my mitochondrial DNA dataset for two populations (N = 11 and N = 32), and I'm running into the same issue on both. I find that I have sub 200 ESS for the Posterior, Prior, ExtendedBayesianSkyline, and final Indicator parameter, but very high ESS for all else. I've tried running for longer, and this is consistent even up to 500 million generations. One thing I have noticed is that the traces of these parameters all begin to jump around considerably after a certain time in the run, and this jumping around is consistent with the final indicator fixing to 1. Might anyone have any idea why this is happening? I've not been able to find any leads. I've attached some images of the traces below.

My set up looks like this:

-Mitochondrial genome alignment partitioned into rRNA1, rRNA2, tRNAs, coding_codon1, coding_codon2, coding_codon3. These all have a single linked tree, but unlinked substitution models and clocks.

-Each is using bModelTest for the substitution model

-I have estimated mutation rates for each partition using Beast with a species tree with fossil calibrations and am fixing these rates for each partition to a strict clock (estimate box unchecked on all partitions)

-Tree prior is an Extended Bayesian Skyline Plot

-I've tried running for up to 500 million generations

Thank you for your help!

Zach

Side note: I have also tried running a Coalescent Bayesian Skyline analysis with 5 dimensions and find it runs quickly with high ESS. I am wondering if the Extended analysis really adds much to my data. As I am linking the trees in my partitions, I will still only estimate one tree, and since I am using mitochondrial data, I doubt I would be able to see more than 4 size changes in the data anyway, so this prior does not seem too problematic. Is there something about my set up that might make the Extended plot most appropriate?

[Artem B]

Hello.
Is one of the population presented by the only 11 sequences? What's length of them?
I think it can be not enough informative for so complicated model.

[Artem B]

Also, why do you link partitions trees?

[Zach N]

Hello, Artem!

The one presented in the images only has 11 sequences, though I get the same results for a population of 34 sequences. For both populations, the sequence length is ~15000bp. I am linking the partitions into one tree as they are all on the mitochondrion.

I think you are likely correct about not having enough informative data. If I move down to a standard Bayesian Skyline Plot with 3 size changes, I get good mixing and ESS for the model. The plot is also relatively flat, so I imagine the Extended would have trouble converging on a number of changes.