Doubt on Clock Rate and Calibration Prior - Can I use Lognormal Clock Prior and Normal Distribution Prior on MRCA. Or should I use Same Distribution on both the Priors?

141 views
Skip to first unread message

VijayaSai A

unread,
Apr 2, 2024, 2:41:20 AM4/2/24
to beast-users
Hi,

I am carrying out evolutionary studies using BEAST v2.7.6. 

The data has seven partitions. The 'Site Models' were 'unlinked' across all the seven partitions whereas the 'Clock models' and 'Tree' were linked across all the partitions. The Clock Model selected was 'Optimized Relaxed Clock' under the 'Clock Model' tab.

In the 'Priors' tab, 'Yule Model' was selected as 'Tree.t:Tree' & by default, 'Log Normal [1.0, 0.2]' was selected under 'ORCRates.cClock' (Optimized LogNormal Relaxed Clock).

My doubt here is, I have selected a LogNormal Relaxed Clock. Whereas I have chosen 'Normal' distribution for MRCA prior. Is this correct? Or should I change the distribution to LogNormal? 

I am attaching the Screenshots. Kindly go through it.
Please help.
Eagerly awaiting for your reply.
With Regards,
VijayaSai  

Alexei Drummond

unread,
Apr 2, 2024, 2:28:14 PM4/2/24
to beast-users
Hi VijayaSai,

The form of the MRCA prior does not have to have any particular correspondence to the model of molecular clock that you choose. However I would always recommend using a log-normal prior instead of a normal prior for all internal node calibrations. Node ages are scale parameters that must be positive, so log-normal is the appropriate prior for such a parameter.

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 view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/87242ec1-b3a9-4b93-aa72-c647d6b628acn%40googlegroups.com.

Farman, Mark L.

unread,
Apr 3, 2024, 1:11:33 PM4/3/24
to beast...@googlegroups.com
Dear Alexei,

I am not sure I understand your reasoning below. I am reanalyzing data from another research group and ran a Coalescent Extended Bayesian Skyline analysis using their parameters which had selected model “None” for their MRCA prior calibrations. Using 10 separate runs, I found the oldest MRCA.date ranged from about -2000 to -4405. In response to your suggestion to always use log Normal distribution for internal node calibrations (I assume you mean MRCA prior here), I tried re-running the analysis using log Normal. Although the analyses converged on a similar substitution rate, all of the MRCA.date values are indeed now positive (between 1 and 2). This suggests to me that log Normal is also enforcing MRCA dates to be positive. This is a problem because MRCA dates can be - and probably most of the time should be - negative.

Please correct me if I am misunderstanding something.

Thanks,

Mark



On Apr 2, 2024, at 2:27 PM, Alexei Drummond <alexei....@gmail.com> wrote:

You don't often get email from alexei....@gmail.com. Learn why this is important
CAUTION: External Sender

Vijaya Sai

unread,
Apr 4, 2024, 12:10:24 AM4/4/24
to beast...@googlegroups.com
Hi Dr. Alexei,

Thank you so much for clarifying the doubt.
Regards,
VijayaSai 

Reply all
Reply to author
Forward
0 new messages