Multiple calibrations - uniform distributions resulting in BEAST crashing

[Rodolfo Probst]

Dear all,

I am trying to set a BEAST analyses using two calibration priors based on secondary calibrations. I tried different distribution schemes but was recommended to - considering the uncertainty of secondary calibrations - be conservative and perhaps give it a try using uniform distributions for both calibrations.

I set one for the root (92-127 Ma) and one for an internal node (50-103 Ma). I've been running different schemes for those two calibrations (e.g., root = normal; internal node = lognormal, both normal distrib., both lognormal) without any issues, but I keep receiving the "Fatal exception: Could not find a proper state to initialise. Perhaps try another seed" when trying to run BEAST with both calibrations as uniform priors. I checked Remco's post on what could be causing it (link here) but after trying to change the seed number, running both on command line and also on CIPRES, I keep getting the same fatal error.

I fixed the tree topology, and there is no conflict with monophyletic taxa (i.e., the clade specified in the calibration is monophyletic and doesn't contain extra taxa) or a single taxon MCRA prior. I was wondering if two MRCA priors are somehow not compatible and if you could provide any feedback on this.

Thank you a lot!

Rodolfo

[Remco Bouckaert]

Hi Rodolfo,

When using a random starting tree, non-overlapping clades will be reported. If you are absolutely sure that the clades have been defined correctly (in my experience it is quite easy to accidentally leave one taxon out), there is still the possibility that the ages in the Newick starting tree are too incompatible with the calibrations and you may need to scale the tree to ensure the calibrations do not run into numerical issues.

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/6affb91d-2d6c-49be-91dc-ad94b78bb79cn%40googlegroups.com.

[Rodolfo Probst]

Hi Remco,

Thank you for your feedback! Not sure if it makes sense, but I turned off all tree search parameters (already ran ML and Bayesian analyses using other software), and my calibrations are as following: root = all taxa; node calibration = all taxa minus one taxon (prior corresponds to an attempt at calibrating the age of a tribe that includes my group of interest and its sister taxon). So I think I set the clades accordingly. But when you said that there is a possibility that the ages in the Newick starting tree are incompatible, I am not sure how to solve this issue. Could you provide some input on how to scale the tree? Or should I only provide the min. age for those priors for the uniform distribution?

Thanks again,

Rodolfo

[Remco Bouckaert]

Hi Rodolfo,

Do you have branch lengths specified in the Newick tree? If only the topology, but no time information is available in the Newick tree, the tree parser generates a very small tree.

If there is time information, there is a scale-input on the TreeParser, so setting a value of say scale=“10” should result in a tree that is 10x as high as the input tree. Just pick an appropriate scale factor.

Cheers,

Remco

To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/411dc2e9-284f-4369-8ff8-8ef6ea00f968n%40googlegroups.com.

[Santiago Sánchez]

Hi Rodolfo,

I wrote a tutorial on my website on how to build a compatible starting tree with Chronos in R. Perhaps it might help: 

https://sites.google.com/u/0/d/1DR2NIgaGmtk89dXjPFn_GF4F1lmSfYW2/p/1VgBGjWepCNCEnFbaILRDkqkNfDkEw3rE/preview

Cheers,

Santiago

To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/C56460E9-C69C-4364-9E11-4D4C1CBE2E00%40gmail.com.

[Santiago Sánchez]

Sorry, this is the right link: 

https://sites.google.com/view/sanchez-ramirez-s/programming/tutorial3

[Rodolfo Probst]

Hi Santiago,

Thank you for your help and for sharing your tutorial - this is great! The thing is that I am already using chronos to quickly transform the output of my IQ-Tree runs into ultrametric - but I didn't add any calibrations while running it. So I guess I could add both uniform calibrations using R like you did in your tutorial and give it a try!

Thanks!

Rodolfo

[Rodolfo Probst]

Hi Remco,

Thanks for the feedback! I do have branch lengths specified in the Newick tree. I obtained the input tree with an IQ-Tree run and transformed it to ultrametric by performing a quick dating analysis with the chronos function in the R package ape, and used that ultrametric tree for my BEAST runs. I kept the scale factor as the default "1.0", but will follow your recommendation and try to pick an appropriate scale factor.

Thanks,

Rodolfo

[Rodolfo Probst]

Hi Santiago,

Thanks again for sharing your tutorial. I was able to add the calibrations to my input tree and BEAST is running!

Best,

Rodolfo

[Santiago Sánchez]

Good to know it worked! --Yes, uniform calibrations are unfortunately tricky to get to work on BEAST.

To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/e9a19d24-92aa-48b0-b204-63518aab7cf1n%40googlegroups.com.