priors for clock rate in BEAST2 (narrow HPD intervals in output)

[Laura Cespedes]

Hi all!

I just started using BEAST2 and have some questions on the differences with BEAST when setting the priors for the clock rate. I am choosing a relaxed clock (log normal) in the 'Clock Model' tab and here I am setting the rate at 0.0125. However, I can't choose estimate. Therefore, in the priors tab I can only set a prior for the standard deviation of the rate, not the mean. I read in the answer of other post ("Why I can not see the ucldMean.c in the priors panel?") that this is because I do not have 'additional information to inform the timing of the tree', which make sense to me. However, when I compare the same tree obtained with BEAST and BEAST2, the HPD intervals for node heights are much narrower for the BEAST2 output. And I think this is misleading. Do someone have any thoughts on this? After changing the standard deviation prior in multiple runs, the mean of this parameter ends up converging at similar values, but also appreciate any advice for setting this prior. Any insights and opinions are very much appreciated. Thanks all in advance.

Best,

Laura C.

[Alexei Drummond]

Dear Laura,

You need either (i) time-stamped data or (ii) node calibrations in order to estimate the molecular clock rate. 

I assume you don’t have either of those? Besides fixing the rate, the final option you have is to specify an informative prior on the rate. Is that what you are trying to do?

Cheers

Alexei

--
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 https://groups.google.com/group/beast-users.
For more options, visit https://groups.google.com/d/optout.
<Captura de pantalla 2018-09-10 a las 7.59.13 p.m..jpg>

[Laura Cespedes]

Dear Alexei,

Thank you very much for your answer. I do not have either of those. What I do have is a reasonable value for the clock rate, and I want to set this rate but allowing some variation. That is why I am not choosing strict clock, but relaxed log normal instead. I am not quite sure if I am specifying an informative prior on the rate by doing this, because I am not setting a prior for the rate itself only its standard deviation. Is what I am doing correct? Or should I just fix the clock rate? Thank you very much again for your help. 

Best,

Laura C.

[Alexei Drummond]

Dear Laura,

The relaxed clock does not model uncertainty in the mean rate of evolution per se. It models real differences in the rate of evolution across branches. Those individual rates are also uncertain so that the rate along each branch has its own posterior distribution. Additionally you might be uncertain both about the mean and the variance of those rates, so that you would try to estimate the mean molecular clock rate and the spread of the lognormal distribution that governs them.

Estimating the mean molecular clock rate without time-stamped sequence data or calibrations is possible by unchecking “Automatic set clock rate” in the “Mode” menu. The reason that we don’t allow this by default is that it is easy to construct an analysis that is not valid. After you have unchecked this you are free to check “estimate” for the clock rate. However you then must go to the “Priors” panel and put an informative prior on the ucldMean parameter. I strongly recommend a log normal prior and not a normal prior since the parameter is a scale parameter and so only priors that go to zero at zero make sense. I have observed bad behaviour when a normal prior is put on the rate.

Cheers

Alexei

[Laura Cespedes]

Dear Alexei,

Thank you very much for your help! I think I will try unchecking “Automatic set clock rate”.

Best,

Laura