low ESS for prior and posterior distribution

[RH]

Hello,

I've just started using BEAST and I'm having some problems, I hope you
can help me.

I want to date the divergence time for a node in my phylogenetic tree.
My tree is well resolved, and has high branch support. I'm currently
using a concatenated fragment of mtDNA, containing cytb, coi and cr.

I did a first run, no codon partions and GTR (I+G), with fixed
molecular clock (rate 1.0), a coalescent tree prior with constant
size, and using the default values for the priors and operators.

This ran fine, the trace files were good, ESS was good for all
values.

I've ran another version of the same file, but with a relaxed
molecular clock and, again, I've got good results, including a
ucld.stdev<1.

However, when I partion the data according to codons and non-coding
region I get very low ESS for the prior and posterior distribution,
with a likelihood of 0 during the run.

I've tried to run each mtDNA fragment alone, and the CR does well, but
not cytb and coi (which, again, I asked to partion into codons).

I've also tried to change the tree prior to YULE process, change prior
distributions and operator values to no end.

I don't really know what to do next, so any help is most welcome.

Thanks.

Romina

[mountainmanjared]

Hi Romina,

This sounds somewhat similar to a problem I've had in *BEAST. For me,
the problem came from over-parameterizing the models of evolution for
each partition. Look at your files in Tracer and see the parameters
that have low (red) ESS values; if some of your nucleotide
substitution parameters for each partition are low (filename.CP#.at
[or .ag, etc.]), there is probably not enough information within each
partition to help the analysis converge. Try keeping your data
partitioned by codon, but giving the partitions a simpler model, like
the HKY model (even if your model testing program told you
otherwise). This changed my ESS values on the posterior and prior
from around 20 to well over 800!

Good luck,
Jared