Problems with clocks and parameters
52 views
Skip to first unread message
Gonçalo Costa
Oct 26, 2017, 12:01:44 PM10/26/17
to beast-users
Hello everyone!
I have a multilocus dataset with 4 partitions (2 nuclear + 2 mit). I am performing tmrca estimates with *BEAST as implemented in v.1.8.6, with a taxaset of 2 spp. + other species clumped under a single large clade (since these are the entities that the nuclear dataset can resolve as monophyletic). The clock estimates I am using are from Marshall et al. 2017 (https://doi.org/10.1093/sysbio/syv069).
I am having an issue with the trmca estimates I am obtaining for each separate mitochondrial and nuclear dataset, and the joined dataset (mito+nuc). The tmrca mean for the same set is, for example, 26.9 Ma for the nuclear, 10.94 Ma, for the mitochondrial and 7.0 Ma for the mit+nuc dataset. Is there a reason for such a discrepancy in the estimates? And why does the combined dataset have a tmrca much younger than the separate datasets? Both tmrca for mito and mito+nuc show a lognormal distribution, while the nuclear shows a distribution abbuting 0, as shown in attachment. All parameters in the log file have ESS well over 200.
Also, in some preliminary runs I have found that the ucld.mean of a partition would sometimes flatline and stop sampling. These runs were ran in triplicate (nruns=600x10⁶) and in only two the MCMC sampling had stopped in different times. see fig2 in attachment. What could cause the estimation to stop? I have solved this problem by limiting the ucld.mean to 1000x larger or smaller than the mean, but the question still remains.
Thank you in advance, this group helped solve many other questions before!
Gonçalo
I have a multilocus dataset with 4 partitions (2 nuclear + 2 mit). I am performing tmrca estimates with *BEAST as implemented in v.1.8.6, with a taxaset of 2 spp. + other species clumped under a single large clade (since these are the entities that the nuclear dataset can resolve as monophyletic). The clock estimates I am using are from Marshall et al. 2017 (https://doi.org/10.1093/sysbio/syv069).
I am having an issue with the trmca estimates I am obtaining for each separate mitochondrial and nuclear dataset, and the joined dataset (mito+nuc). The tmrca mean for the same set is, for example, 26.9 Ma for the nuclear, 10.94 Ma, for the mitochondrial and 7.0 Ma for the mit+nuc dataset. Is there a reason for such a discrepancy in the estimates? And why does the combined dataset have a tmrca much younger than the separate datasets? Both tmrca for mito and mito+nuc show a lognormal distribution, while the nuclear shows a distribution abbuting 0, as shown in attachment. All parameters in the log file have ESS well over 200.
Also, in some preliminary runs I have found that the ucld.mean of a partition would sometimes flatline and stop sampling. These runs were ran in triplicate (nruns=600x10⁶) and in only two the MCMC sampling had stopped in different times. see fig2 in attachment. What could cause the estimation to stop? I have solved this problem by limiting the ucld.mean to 1000x larger or smaller than the mean, but the question still remains.
Thank you in advance, this group helped solve many other questions before!
Gonçalo
Reply all
Reply to author
Forward
0 new messages