examl bootstopping not converging

[ctda...@gmail.com]

Hi everyone

I'm running a 2.8 million bp alignment with 167 taxa on examl with 1341862 alignment patterns generated from RADseq.

I'm bootstrapping the support values - I'm now up to 275 trees and still the bootstopping step is not showing convergence. Whether I generate the bootstrap values from 100, 200 or 275 bootstrap replicates the support values are almost identical, so adding more replicates does not seem to be improving the bootstopping. I am puzzled how to interpret this - many of my species have 100% support while only a couple are around 70% which could well be biologically plausible.

Any ideas why it's not converging?

Many thanks

Clive Darwell

[Alexey Kozlov]

Hi Clive,

how do you check for convergence? Are you using a posteriori bootstopping with RAxML "-I autoMRE" option (you should)?
E.g.:

raxmlHPC-SSE3 -n convergenceTest_autoMRE -z bunch.nw -I autoMRE -m GTRGAMMA -p 13477 -B 0.03

If so, does the convergence criterion value improves as you add more bootstrap. You should see something like this in
the RAxML_info file:

# Trees Avg WRF in % # Perms: wrf <= 3.00 %
50 3.27 401
100 2.85 479
150 2.32 822
200 1.87 976
250 1.63 986
300 1.38 1000
Converged after 300 replicates

It might well be that 275 bootstrap replicates are not enough to reach convergence, and you have to run more.

Best,
Alexey

> --
> You received this message because you are subscribed to the Google Groups "raxml" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to raxml+un...@googlegroups.com
> <mailto:raxml+un...@googlegroups.com>.
> For more options, visit https://groups.google.com/d/optout.

[Alexandros Stamatakis]

also, sometimes a bootstrap not converging indicates that the likelihood
surface may have two distinct peaks, happened to me when I was preparing
this chapter here:

http://onlinelibrary.wiley.com/doi/10.1002/0471250953.bi0614s51/abstract

I had randomly selected an empirical test dataset from my collection and
the BS wouldn't converged. I then checked where that came from and found
that it was a "classic" dataset used for testing Bayesian inference
codes and had two distinct areas of high posterior probability ....

finally, I wouldn't expect radseq datasets to have very stable support
anyway due to the missing data patterns

alexis

--
Alexandros (Alexis) Stamatakis

Research Group Leader, Heidelberg Institute for Theoretical Studies
Full Professor, Dept. of Informatics, Karlsruhe Institute of Technology
Adjunct Professor, Dept. of Ecology and Evolutionary Biology, University
of Arizona at Tucson

www.exelixis-lab.org

[ctda...@gmail.com]

Dear Alexei

Thanks for your response. Interesting. Yes, I was using the -I autoMRE option. I shall try with more bootstraps.
Also, if I plot the likelihood surface would it be clear that there is more than one peak?

Best wishes

Clive

[Alexandros Stamatakis]

On 17.01.2017 00:36, ctda...@gmail.com wrote:
> Dear Alexei
>
> Thanks for your response. Interesting. Yes, I was using the -I autoMRE
> option. I shall try with more bootstraps.

you should not give up until you reach 1000 replicates

> Also, if I plot the likelihood surface would it be clear that there is
> more than one peak?

I guess not, I'd compute 50 ML trees on the original data and then look
at their likelihoods and maybe compute the pair-wise RF distances, that
might yield some insight, you may also want to do MDS plots using that
data (likelihoods and RF-distances)

alexis

>
> Best wishes
>
> Clive
>
> On Monday, January 16, 2017 at 4:27:30 PM UTC+9, ctda...@gmail.com wrote:
>
> Hi everyone
>
> I'm running a 2.8 million bp alignment with 167 taxa on examl with
> 1341862 alignment patterns generated from RADseq.
>
> I'm bootstrapping the support values - I'm now up to 275 trees and
> still the bootstopping step is not showing convergence. Whether I
> generate the bootstrap values from 100, 200 or 275 bootstrap
> replicates the support values are almost identical, so adding more
> replicates does not seem to be improving the bootstopping. I am
> puzzled how to interpret this - many of my species have 100% support
> while only a couple are around 70% which could well be biologically
> plausible.
>
> Any ideas why it's not converging?
>
> Many thanks
>
> Clive Darwell
>