MCMCTree: Is it problematic to use a root node calibration only for relaxed dating?

[Michael Branstetter]

Dear Group,

I recently performed relaxed dating on a phylogenomic dataset using BEAST2 and MCMCTree and recovered oddly discordant results that I cannot explain. For each analysis I assigned a single node calibration to the root node and used a normal distribution for the prior. As far as I can tell, the settings between programs were as similar as possible. I used a relaxed lognormal clock, birth-death tree prior, and GTR+G sequence model. For BEAST2 I used 500 clocklike loci, and for MCMCTree I used the same 500 loci and all loci. For both MCMCTree runs, I first calculated the Hessian matrix (usedata=3), and then performed dating using the approximate method (usedata=2).

The BEAST2 results match my expectations. The root node is old and their is a long branch leading up to the much younger in-group. This pattern appears similar to the undated phylogram. In contrast, both of the MCMCTree trees push the ingroup nodes much closer to the root node. In fact the nodes look roughly evenly distributed across time. The result is the same for all loci or just the 500 clock loci.

I have had more congruent results when using multiple calibration points, so I am wondering if the difference here has something to do with using a root-node calibration only. I am not sure why that would be and why the MCMCTree result would differ so much from the BEAST result. Is there a setting that I am forgetting to consider that could have a major affect on node heights? The BEAST2 result seems more realistic to me, but it is hard for me to explain why it is performing better. I would appreciate any ideas that people have.

Best,

Michael

[Sishuo Wang]

Hi Michael,

Can I ask which clock model  you used? My limited experience is that there is a possibility that some different results could arise from the use of different clocks (IR, AR or STC). You may wish to have a look at Betts et al. 2018 (Fig. 1) or Wang & Luo 2020 (Fig. 2A).

Sishuo

[Ziheng]

i am not sure what's happening.  can you paste your mcmctree control file here?

are you treating the 500 loci as one partition?

ziheng