Calibration points not taking effect in MCMCTree (PAML 4.9)

[guo]

Hi Kostas,
I hope you are doing well.
I am writing to ask for your advice regarding a calibration issue I encountered when running MCMCTree in PAML version 4.9.
In my analysis, I specified two calibration points in the tree file (as shown in the attached tree), and the MCMC run completed successfully. However, when I visualized the resulting output tree (.tre) in FigTree, it appears that the calibration constraints were not reflected in the divergence time estimates, as if the calibrations did not take effect.
Thank you very much for your time and help.
Best regards

[Sandra AC]

Dear Guo,

Thanks for your message! I am not entirely sure I understand what you mean by "the calibration constraints were not reflected in the divergence time estimates". 

Did you run a first analysis without sequence data to compare the marginal densities with the calibration densities you specified in the tree file? You may do this with custom in-house R scripts or by loading your "mcmc.txt" file (output file with samples collected during the MCMC when the target distribution is the prior) on Tracer to visualise how the marginal densities (oftentimes referred to as "effective prior") inferred by MCMCtree align (or not) with your user-specified calibration densities. This is an important step that should not be skipped as it will help you understand whether you need to adjust your user-specified calibration densities prior to using the sequence data for timetree inference. You may want to check the PAML Wiki for the MCMCtree program for more details about analyses without sequence data, specially sections "Overview" and "How to visualise your calibration densities?".

Hope this is somewhat useful but, if you have further questions or need additional clarifications, please do not hesitate to get back!

All the best,
Sandy

[Sishuo Wang]

Dear guo,

Do you mean that the posterior ages are largely outside your calibration? if so, it'd be great if you could share the input and outputs such that sandy, i or others can more easily help you solve the Q. thx!

best,

sishuo