modeltest with partitioned dataset + heterotachy version

[Karen Meusemann]

hi all,

sorry for this probably stupid question - if I want IQtree to estimate the best model for each partition, how my partitionfile (nexus) should look like then? (or will -m TEST or -m TESTNEW "overrule" something given in the partition file (nexus format) ? Moreover I was wondering if I can also to this with the "heterotachy" version (1.4.3 heterotachy) having partitions ready ?

Thanks Karen

[Bui Quang Minh]

Hi Karen,

On Nov 8, 2016, at 6:28 PM, 'Karen Meusemann' via IQ-TREE <iqt...@googlegroups.com> wrote:

hi all,

sorry for this probably stupid question

not stupid at all ;-)

- if I want IQtree to estimate the best model for each partition, how my partitionfile (nexus) should look like then? (or will -m TEST or -m TESTNEW "overrule" something given in the partition file (nexus format) ?

by default -m TEST (or -m TESTNEW) will overrule all models specified via “charpartition" command in the NEXUS file. In fact it suffices to just have “charset” command in NEXUS file. IQ-TREE will then find best model for each partition and disregard charpartition.

Moreover I was wondering if I can also to this with the "heterotachy" version (1.4.3 heterotachy) having partitions ready ?

does it mean, you want to have +H (rate heterotachy model) for each partition? This is not possible, as it can be well over-fitting with too many parameters (for example, +H2 adds 2*(2n-3) branch length parameters to estimate, where n=#taxa). Right now, +H is meant to be used with the whole length phylogenomic alignment, and not to combine with partition models.

FYI: we are now modifying the heterotachy model to work with other mixture models, for example LG+C10+H4. I will let you know once that happens.

Minh

Thanks Karen

--
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.

--

Bui Quang Minh

Center for Integrative Bioinformatics Vienna (CIBIV)

Campus Vienna Biocenter 5, VBC5, Ebene 1

A-1030 Vienna, Austria

Phone: ++43 1 4277 74326

Email: minh.bui (AT) univie.ac.at

[Karen Meusemann]

Hi Minh,

thanks for the answer (I answered last week alreay but got lost somehow.
Thanks for your comments. ok - I actually was thinking to try this because of the heterogeneity acros the tree (so lineages) - so if it only to make sense to apply here this on an unparetitioned dataset I probably will for now not try this version because I probably can't compare what is better, partitioned with all kins of models or unpartitioned mixtuer model). How is heterogeneity along the tree (taxon sampling) addresse in the heterotachy version (or is it at all?)

Thanks
Karen

[Bui Quang Minh]

Hi Karen,

On Nov 14, 2016, at 11:06 AM, Karen Meusemann <kmeus...@googlemail.com> wrote:

Hi Minh,

thanks for the answer (I answered last week alreay but got lost somehow.
Thanks for your comments. ok - I actually was thinking to try this because of the heterogeneity acros the tree (so lineages) - so if it only to make sense to apply here this on an unparetitioned dataset I probably will for now not try this version because I probably can't compare what is better, partitioned with all kins of models or unpartitioned mixtuer model).

How is heterogeneity along the tree (taxon sampling) addresse in the heterotachy version (or is it at all?)

I think it is right now controversial when to use partition or mixture models…  several experts discussed about this in http://phyloseminar.org, see the recent series about heterogeneous substitution by Andrew Roger, Nathan Whelan and Nicholas Lartillot. I would personally recommend that one should apply both partition and mixture models to gain more confidence if the results agree. 

About the heterotachy version: we are still experimenting this model, thus more analysis and verification are needed. We will let you know when it is more ready to use.

Cheers, Minh