Truncating distributions and infinity issues.

715 views
Skip to first unread message

JHallas

unread,
Mar 26, 2015, 3:10:20 PM3/26/15
to beast...@googlegroups.com
Hello BEAST-users,

I keep encountering an issue when i try to truncate the distribution of my tree priors. 

Both calibration dates are close to zero, 5.5MYA and 5.45MYA.  I have read the having dates close to zero can have a large effect on the treeMoldel.rootHeight.  When i run my analysis without truncating the prior distribution (normal distribution) the 
treeMoldel.rootHeight  is being estimated as being almost zero (mean: 3.2509E-100).  Deciding to truncate the prior i now get an error that

"Parsing error - poorly formed BEAST file:
Error parsing '<uniformPrior>' element with id, 'null':
Uniform prior uniformPrior cannot take a bound at infinity, because it returns 1/(high-low) = 1/inf"

The only uniform distribution I am using is ucld.mean prior. Initial value .033, upper 1.0E100, Lower 0.0. Can this be the reasin i am encountering this error?
If so what is the best way to resolve the parsing error?

Thanks

-Joshua Hallas

Santiago Sánchez

unread,
Mar 26, 2015, 5:36:00 PM3/26/15
to beast...@googlegroups.com
Hi Joshua,

What prior(s) are you using for your calibrations? Are these based on fossil data? If so, you can try with soft-bound prior distributions such as lognormal, gamma, or exponential. In any case, you regularly set as an offset your minimal clade (fossil) age. The upper soft-bounds of the distribution may de determined using "reasonable" biological data (i.e. this clade cannot be older than X Ma based on such and such..). However, If your dates are, for instance, mean divergence times from other molecular dating studies, your could use a normal distribution with a standard deviation matching the 95% confidence interval. In general, I would try to stay away from uninformative (improper) priors such as uniform distributions. For the ucld.mean, you can try with the CTMC reference prior, which was design to deal with parameters for which your have little or no information.

Cheers,
Santiago


--
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 post to this group, send email to beast...@googlegroups.com.
Visit this group at http://groups.google.com/group/beast-users.
For more options, visit https://groups.google.com/d/optout.



--
Santiago Sánchez-Ramírez
Department of Ecology and Evolutionary Biology, University of Toronto
Department of Natural History (Mycology), Royal Ontario Museum
100 Queen's Park
Toronto, ON
M5S 2C6
Canada

Joshua Hallas

unread,
Mar 26, 2015, 7:02:20 PM3/26/15
to beast...@googlegroups.com
Hi Santiago,

I am using geological dates (Bering Strait and the Baja CA formation) to calibrate a Lognaormal uncorrelated relaxed clock.  I initially set each tmrca prior to normal distribution with 95% confidence intervals based on the estimate for the geological event ie opening of the Bering Strait 5.5-5.4Ma.  However, reviewers wanted to see a larger stdev because of the large uncertainty around geological dates. In addition, I also recovered very narrow 95% HPD for calibrated nodes, which raised questions about the distributions range being too narrow.  After setting a larger stdev and excluding the outgrouping i started encountering these issues. 

I have just ran an analysis using CTMC as you suggested and still i am unable to truncate the tmrca priors and the treeMoldel.rootHeight mean=3.5015E-100.  Is there a reason why the
treeMoldel.rootHeight mean could be so small?  I ran the same analysis with the outgroups included and treeMoldel.rootHeight mean=47.8417.  In both instances i did not set a distribution to the treeMoldel.rootHeight prior.  There is no fossil record for this group, nudibranchs, so constraining the root is problematic. 

-Josh

--
You received this message because you are subscribed to a topic in the Google Groups "beast-users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/beast-users/T6QTxfzDozk/unsubscribe.
To unsubscribe from this group and all its topics, send an email to beast-users...@googlegroups.com.

Santiago Sánchez

unread,
Mar 26, 2015, 8:07:36 PM3/26/15
to beast...@googlegroups.com
OK. What values are you using now for your normal calibration(s) (i.e. mean and stdv)? In theory, a single source of time information should be enough to perform a molecular dating analysis. So for instance, if you are using a TMRCA prior, don't use an additional molecular clock rate prior. With this I also mean that you don't need to constrain the root of the tree.

Other questions:
Did the run converge/mixed well (ESS > 200)?
Are you using values in Ma (for instance, 5 ± 1 Ma)?
How many tips does your tree have?
What tree model prior are you using (coalescent / speciation)?
BEAST version?
Are you getting the same error message with the new runs?

Sorry to have more questions instead of answers...

Santiago

Joshua Hallas

unread,
Mar 26, 2015, 10:24:29 PM3/26/15
to beast...@googlegroups.com
Santiago.

I have 3 TMRCA priors.  Once aspect of this research is to see how different calibration scenarios change the divergent time estimates.  The analysis works perfectly when all three calibrations are used.   When the much older date (95-105Ma) is excluded I encounter rootheight issues.

Did the run converge/mixed well (ESS > 200)? my ESS values are > 2000
Are you using values in Ma (for instance, 5 ± 1 Ma)? Yes.  I took the mean of the geological date and set the stdev to cover the range.  ie 95-105Ma = mean:100 stdev 2.45
How many tips does your tree have? 11
What tree model prior are you using (coalescent / speciation)? Speciation: Yule Process.  I've tried other models and I still get the same issues regarding the rootheight
BEAST version? v1.8.1.  I've also encoutered the same issue using BEAST 2 and BEAST 1.7.4
Are you getting the same error message with the new runs? I get error messages only when i truncate the priors for tmrca.  When i don't truncate the analysis runs however i get very small rootheight.

Thanks for the help Santiago.  I'm trying to figure this issue out for months now.



Santiago Sanchez

unread,
Mar 27, 2015, 8:24:07 AM3/27/15
to beast...@googlegroups.com
Joshua,

I'm guessing that when you use the youngest calibration (5 Ma?) only the issue arises, right?

Anyway, its bizarre that you would get such young ages for the root of tree even when having one 5Ma calibration. What is the posterior of the 5Ma calibrated node after the analysis? And again, what prior distribution are you using for that node?

If you like you could send me your XML and I can have a quick look.

Cheers,
Santiago

Santiago Sanchez-Ramirez
Ecology and Evolutionary Biology, University of Toronto
Natural History (Mycology), Royal Ontario Museum
100 Queen's Park
Toronto, ON
M5S 2C6
Canada

Joshua Hallas

unread,
Mar 31, 2015, 9:30:26 PM3/31/15
to beast...@googlegroups.com
ya i encounter issues when i take out the outgroups and use only the 5Ma calibration.  The posterior is 1, and the distribution is normal.  Its weird alright.  A run just finished where i changed a few things.  Hopefully I fixed the issues. 


-Josh

sshweta

unread,
Feb 5, 2018, 5:04:23 AM2/5/18
to beast-users
H

sshweta

unread,
Feb 5, 2018, 5:05:47 AM2/5/18
to beast-users
HAVE you put all 3 TMRC simultaneously or one by one the combined??
What is the format to set the TMRS as I have also 3 TMRC but dont know how to use
Please reply asap..


On Friday, March 27, 2015 at 7:54:29 AM UTC+5:30, JHallas wrote:
Reply all
Reply to author
Forward
0 new messages