Theoretical basis for bootstrap analysis "failure to converge"

[Zachary Griebenow]

Greetings; this is a theoretical rather than operational query. Whenever I run IQ-Tree, I do so with ultra-fast bootstrapping under the -bnni option (just on the off-chance that there is a severe model violation). I receive an error message stating that the bootstrap analysis failed to converge, and a suggestion that I increase the number of iterations to compensate for this. Upon researching the question, I have learned that when using -bnni option, this error can be safely ignored. However, I would like to understand why. What is meant by "convergence" in the context of maximum likelihood? And why is it permissible to ignore this error message when running IQ-Tree with -bnni, but not otherwise?

Thanks for your time!

Regards, Zach

[Minh Bui]

Hi Zach,

This is a bit long to explain. The UFBoot assesses convergence in the support values as follows. Let’s say at the search iteration 200, we gather the set of all bootstrap trees: S_200, which induces a set of splits (branches) and their occurence frequencies among the boostrap trees. The occurence frequencies are compared with those of S_100 (from iteration 100): if the correlation coefficient is >0.99, we say that UFBoot converges and stops.

Now with the new option -bnni in UFBoot2, the set S_200 is “refined” (with NNI, see details in https://www.ncbi.nlm.nih.gov/pubmed/29077904), resulting in a modified set of trees S’_200. The WARNING is not correct, because we computed the correlation between S_100 and S’_200, which might be <0.99 and thus wrongly reported as not convergence. This is the behaviour up to version 1.6.10.

Therefore in the latest version 1.6.11, we fixes this by now reporting the correlation between S_100 and S_200. Thus, the WARNING should not appear anymore, if the bootstrap values indeed converged.

Hope that answers your question.

Cheers,

Minh

--
You received this message because you are subscribed to the Google Groups "IQ-TREE" group.
To unsubscribe from this group and stop receiving emails from it, send an email to iqtree+un...@googlegroups.com.
To post to this group, send email to iqt...@googlegroups.com.
Visit this group at https://groups.google.com/group/iqtree.
To view this discussion on the web visit https://groups.google.com/d/msgid/iqtree/0f61891d-2fb9-4436-b41a-7d172d6b772c%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Zachary Griebenow]

Good morning Dr. Minh, thank you; that was exactly the sort of explanation that I was looking for. I will let you know if I am in need of any further clarification regarding this warning from UFBoot.

Regards, Zach

To view this discussion on the web visit https://groups.google.com/d/msgid/iqtree/5567A329-F649-4047-9B09-54FD631305B6%40univie.ac.at.

[Liu Jia]

Hello Minh,

I am now using the IQTREE version 2.2.0, and also got the convergence issue (even with 5000 iterations and 0.2 perturbation) as shown below:

"""

NOTE: Bootstrap correlation coefficient of split occurrence frequencies: 0.925
NOTE: UFBoot does not converge, continue at least 100 more iterations
"""

My command looks like this: 

`iqtree2 -s ../0.95_rogues_removed_msa.fa -m TVM+F+R10 -pers 0.2 -nstop 500 -B 1000 --abayes --alrt 1000 -T AUTO -ntmax 15 --nmax 5000 --boot-trees --prefix iqtree2-nmax-5000-pers-0.2-rep1_0.95_rogues_removed_msa_result`

Here are more information about my sequence alignment:

"""

WARNING: 6 sites contain only gaps or ambiguous characters.
Alignment has 1252 sequences with 1231 columns, 920 distinct patterns
777 parsimony-informative, 117 singleton sites, 337 constant sites

"""

I created 10 trees all with 5000 iterations, perturbations from 0.1 to 0.5 (with increment at 0.1), two replicates at each perturbation. And they all failed to converge. Is there anything you would recommend for the next step to try out? Does adding `-bnni` to my commands help? If with `-bnni` the trees still do not converge, can I ignore the convergence warning with IQTREE v2.2.0? 

Thanks,

Jia