EBSP and low ESS for posterior, prior and locus specific EBS parameters

[lauren]

Hello, 

I am doing EBSP analyses in Beast 2.1.3 for three locus datasets and have been getting low ESS for posterior and priors that seems to be due to the ExtendedBayesianSkyline parameters for the individual loci. The posterior for each of the parameters has a sharp peak in the negatives (eg, -150) but a long flat tail that extends into positive numbers. Longer runs does not alleviate the issue. Does anyone know why this might be happening? Alternatively, does anyone know if we can place an upper boundary on these parameters?

Thanks so much for any help!

-Lauren

[Remco Bouckaert]

Hi Lauren,

There is a known issue with the EBSP analyser in v2.1.3. A fix is available by downloading

https://github.com/CompEvol/beast2/releases/download/v2.1.3/beastEBSP.jar

and run the analyser form the command line using

java -cp /path/to/beastEBSP.jar beast.app.tools.EBSPAnalyser

Does this solve the problem?

Cheers,

Remco

--
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 http://groups.google.com/group/beast-users.
For more options, visit https://groups.google.com/d/optout.

[Lauren Chan]

Hi Remco, thanks for your quick reply. The problem is not with the EBSP analyzer, but rather with the actual run itself. When I look at the log file in tracer, the ExtendedBayesianSkyline parameters are positive. I’ve attached a screenshot. Am I correct that these are likelihoods and should be negative? These plots look similar for the other two loci. The likelihood for the entire run has good mixing and converges across runs, but the prior and posterior both have low ESSs. 

Thanks so much for your help, 

-Lauren

You received this message because you are subscribed to a topic in the Google Groups "beast-users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/beast-users/JVzxITBrH5M/unsubscribe.
To unsubscribe from this group and all its topics, send an email to beast-users...@googlegroups.com.

[Remco Bouckaert]

Hi Lauren,

It took a bit of time to consult with Joseph, who wrote the code. The ExtendedBayesianSkyline is a tree prior, and is not normalised, so it can become positive. Tree likelihoods should always be negative though since these are normalised.

Perhaps the low ESSs can be explained by the lack of data, which makes the MCMC explore a much larger space than when the data directs it to a more restricted part of the parameter space. 

Cheers,

Remco

On 7/01/2015, at 7:51 am, Lauren Chan <laure...@gmail.com> wrote:

Hi Remco, thanks for your quick reply. The problem is not with the EBSP analyzer, but rather with the actual run itself. When I look at the log file in tracer, the ExtendedBayesianSkyline parameters are positive. I’ve attached a screenshot. Am I correct that these are likelihoods and should be negative? These plots look similar for the other two loci. The likelihood for the entire run has good mixing and converges across runs, but the prior and posterior both have low ESSs. 

Thanks so much for your help, 

-Lauren

<Screen Shot 2015-01-06 at 10.49.23 AM.png><Screen Shot 2015-01-06 at 10.49.35 AM.png>

[sandy]

Dear Lauren,

 Have you solved your problem?

 I have encounter the same problem: very low ESS on prior , posterior, and likelihood for the EBSP analyses.

 Would you mind give me some suggestion?

 Thank you very much.

Best regards,

Sandy

在 2015年1月7日星期三 UTC+8上午12:52:47，lauren写道：

[lauren]

Hi Sandy, 

Have you tried the newest version of BEAST (2.2.1)? I know that this solved some of the problems I was having with getting the EBSP analyses to run well.

Lauren