Implementation of combined mixture models (e.g. JTT+UL3+G)

[Anthony Redmond]

Dear IQ-tree users/teams,

A) I have a case where the best fit amino acid models for my dataset is JTT+UL3+G, and wished to enquire about its behaviour.

Does this become a 4-matrix model (JTT+buried+intermediate+exposed), or are the three mixtures of UL3 replaced with independent JTT matrices?

I am assuming this is more like the later case; if so, just for clarity, this would then mean that say, JTT+EX3+G  and JTT+UL3+G would be effectively the same model?

B) Similarly, is this behaviour the same for other n-matrix models (UL2,EX2,EX3,EHO,EX_EHO)? With CF4 on the other hand sharing behaving more like C10-60?

Thank you!

Anthony

[Bui Quang Minh]

Hi Anthony,

That’s an interesting question. In fact I never considered such mixture combination before. It is possible but with some change in model specification:

On Nov 15, 2017, at 6:08 PM, Anthony Redmond <anthon...@gmail.com> wrote:

Dear IQ-tree users/teams,

A) I have a case where the best fit amino acid models for my dataset is JTT+UL3+G, and wished to enquire about its behaviour.

Does this become a 4-matrix model (JTT+buried+intermediate+exposed), or are the three mixtures of UL3 replaced with independent JTT matrices?

I am assuming this is more like the later case; if so, just for clarity, this would then mean that say, JTT+EX3+G  and JTT+UL3+G would be effectively the same model?

1. I’m not sure why JTT+UL3+G was the best-fit model, but a simple run with option "-m JTT+UL3+G”, IQ-TREE ignores +UL3 in the model string. Thus it is equivalent to -m JTT+G.

2. Likewise, if you do -m UL3+JTT+G, then IQ-TREE ignores +JTT, resulting in UL3+G model instead. 

3. If you want to really do a 4-class mixture model, then you need to specify this:

iqtree -m 'MIX{JTT,Q1_UL3,Q2_UL3,Q3_UL3}+G' ...

Note that the model string must be rounded by quote. 

For other combination: you can look at the line like this in the log file when running with -m JTT+UL3+G:

Model JTT+UL3+G is alias for JTT+MIX{Q1_UL3:0.484340397,Q2_UL3:0.492780514,Q3_UL3:1.15597274}+G

And then rename the model string the way I described above.

B) Similarly, is this behaviour the same for other n-matrix models (UL2,EX2,EX3,EHO,EX_EHO)?

Yes, the same behavior for other mixture models you listed.

With CF4 on the other hand sharing behaving more like C10-60?

Yes, CF4 has a different behavior, just like C10 to C60.

Hope that helps,

Minh

Thank you!

Anthony

--
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

[Anthony Redmond]

Dear Minh,

Sorry for the delay in my reply and thank you for your help.

Just a thought on your answers:

I'm delighted that IQ-tree allows the user to specify a model like 'MIX{JTT,Q1_UL3,Q2_UL3,Q3_UL3}+G', but it does not seem to be possible to compare the fit of a mixture of this type against a standard JTT+G, or UL3+G model with IQ-tree/ModelFinder. It might be interesting if something akin to this were possible in future versions as some alignment sites could presumably be better modelled in this way. 

Thanks again for your help,

Anthony

[Bui Quang Minh]

Dear Anthony,

We are working towards a more flexible ModelFinder, i.e., allowing to test different kinds of mixture models. This will hopefully come with version 1.6.

Cheers, Minh

[Anthony Redmond]

Dear Minh,

Sounds great! Thanks again,

Anthony