Advice on tree prior for constrained analysis of within and between species sampling

[Nathanael Walker-Hale]

Hi RevBayes community,

I'm writing to ask for some advice on a topology and divergence time analysis I'm trying, which consists of 5 markers (~3000 sites) for 161 samples. The delimitation of the samples is very uncertain, but they probably represent 5-6 species. Of these samples, 21 have genome assemblies (but are represented in this dataset just by the 5 markers), and there is a topology from a phylogenomic analysis of these 21 species which I would like to enforce as a backbone constraint.

I've managed to run a partitioned time tree using a birth-death prior with dnConstrainedTopology and the backbone argument, taking advice from the first part of Landis' TimeFIG tutorial, here: https://revbayes.github.io/tutorials/timefig_dating/. I'm using a secondary calibration for the root age from a prior divergence time analysis, parameterised as a truncated normal. Due to the mixture of sampling, there's not really a good way to estimate the sampling fraction, so I set rho to 1. This analysis is not mixing particularly well, but I can already see that the intraspecific divergences are really quite old (certainly, older than I would expect, e.g. 4 MYA). I'm wondering if this is a consequence of an inappropriate tree prior, either due to the mixture of inter- and intraspecific sampling or the sampling fraction. So my question is twofold: can one sample rho as a stochastic variable, and if so, what is a good prior for this; and secondly, is there a better tree prior for this mix of sampling? (the only information I could find on this second point was Ritchie & Ho, 2017: https://academic.oup.com/sysbio/article/66/3/413/2682285, which seems to suggest not a huge difference relative to coalescent priors, but I still worry that any of these choices: pure birth-death, or pure coalescent, is not appropriate, and a mixture is required).

I'm weary of StarBeast or BPP style models as I suspect the dataset is going to be too big for them - hoping people might have some good thoughts on options.

Cheers!

Nathanael

[Nathanael Walker-Hale]

wary*! An ironic typo...