Model selection for Lie Markov models

[Edward Braun]

Hi --

Is there a simple option (equivalent to -m TEST or -m MFP) to examine a complete set of the Lie Markov models from Woodhams et al. (2015)

When I did the following:

-m TEST -mset 1.1,2.2b,3.3a,3.3b,3.3c,3.4,4.4a,4.4b,4.5a,4.5b,5.6a,5.6b,5.7a,5.7b,5.7c,5.11a,5.11b,5.11c,5.16,6.6,6.7a,6.7b,6.8a,6.8b,6.17a,6.17b,8.8,8.10a,8.10b,8.16,8.17,8.18,9.20a,9.20b,10.12,10.34,12.12

it worked, although it only examined the RY versions of the models with RY, MK, and WS variants.

Obviously, the line above works and I could add in the RY, MK, and WS variants for the models that have those variants. However, I just wanted to make sure I wasn't missing something obvious (and I apologize if I am)

Thanks!

Ed

[Minh Bui]

Hi Ed,

Quick answer: "-mset liemarkov” will test all Lie-Markov models, including RY, ML and WS variants.

However, I recommend: “-m TEST+LM” and ignore -mset command. This will test all usual reversible DNA models as well as Lie Markov models, thus exploring all DNA models available so far.

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.
For more options, visit https://groups.google.com/d/optout.

[jacob...@gmail.com]

As a follow up to this question -- it seems this may not work in 2.1.1. Is there a way to do a model+tree search with -m MFP+LM+MERGE, but restrict only to the suite of lie markov models? Combining with "--mset liemarkov" does not appear to generate the expected restriction.

In another post, Minh indicated that one should not do a partitioned analysis combining reversible and non-reversible models, but it seems that restricting to lie markov models with --mset doesn't behave as expected. 

Best wishes,

Jake

[Minh Bui]

Hi Pauline,

Sorry, I was a bit confused by the attachment you sent with this thread.

For the Lie Markov models: I was not precise with my answer before: in fact the partition model does not support non-reversible models for each partition yet. So all partitions need to have reversible models. 

Minh

On 9 Sep 2020, at 5:35 pm, Pauline Stud <studp...@gmail.com> wrote:

Hi, I've been trying to analyse a alignment with heterogenous composition so wanted to force the use of non-homogeneous model to avoid artefact due to GC compostion bias. I've read that one sould avoid partitioned analysis combining reversible and non-reversible models, but is it safe if we only have non-reversible ones? I tried that way: iqtree -s alignment.phy -p partitions.nex -st DNA -B 1000 -mset liemarkov. Is it the right command? Do I have to pass a treefile or the name of the root? It resulted in several warning like these:

WARNING: evec not invertible

WARNING: Numerical underflow for non-rev lh-branch Noname

And finally this error log:

new_tree_lh: -3654.0248185988   tree_lh: -3195.2806600962

ERROR: phylotree.cpp:2806: virtual double PhyloTree::optimizeAllBranches(int, double, int): Assertion `fabs(new_tree_lh-tree_lh) < max_delta_lh' failed.

To view this discussion on the web visit https://groups.google.com/d/msgid/iqtree/da22e636-aba8-4663-91a3-73a38dace31dn%40googlegroups.com.

[Pauline Stud]

Thank you. What is the best options to use with nonreversible models: indicate the root? provide a rooted starting tree? As they are not reversible I reckon this model needed a specifed root but I didn't find anything particular in the tutorials/website, same with the parameters options (eg: can we specify equilibrium frequency?Gamma distribution?). I tried several option but I'm not sure what's fine. I also noticed ultrafast bootstrap isn't available when you provide a tree. 

[Minh Bui]

You don’t need to provide a tree, IQ-TREE will find the root automatically. 

Minh

To view this discussion on the web visit https://groups.google.com/d/msgid/iqtree/279faeec-4f8f-41cd-9cfc-bc2e7a6edf4fn%40googlegroups.com.