Assigning different nucleotide substitution models to different partitions

[Hannah Miller]

Dear Alexis,

I was wondering if it is possible to define different nucleotide substitution models to different genes in my partitioned dataset using RaxML.
My dataset consists of the following genes:

DNA, 16S = 1-1483
DNA, ITS = 1484-1976
DNA, 23S = 1977-2626

Modeltest suggested to use GTR+G+I for the 16S and 23S partitions and K80+I for the ITS partition.
Since I would like to compare the ML tree topology with a bayesian inference-estimated tree (MrBayes lets me specify the different nucleotide models) I would like to know whether there is any way in RaxML to specify the equivalent models?
As far as I understood I can only define different GTR models as a command in combination with my partition file (e.g. -m GTRGAMMAI -q partition). The GTR model would thus be assumed for all the partitions however. Is there any way to include the different models in the partition text file, such as described for the different protein models?

I would be very grateful for advice!

Cheers Hannah

[Alexandros Stamatakis]

Dear Hannah,

> I was wondering if it is possible to define different nucleotide
> substitution models to different genes in my partitioned dataset using
> RaxML.
> My dataset consists of the following genes:
>
> DNA, 16S = 1-1483
> DNA, ITS = 1484-1976
> DNA, 23S = 1977-2626
>
> Modeltest suggested to use GTR+G+I for the 16S and 23S partitions and K80+I
> for the ITS partition.
> Since I would like to compare the ML tree topology with a bayesian
> inference-estimated tree (MrBayes lets me specify the different nucleotide
> models) I would like to know whether there is any way in RaxML to specify
> the equivalent models?

Unfortunately, this is not possible. Please also check the discussion of
using the combination G+I in the RAxML manual, this model is
non-identifiable.

I personally don't think that the DNA substitution models matter that
much (in contrast to Protein models), so using GTR+G should mostly be
fine, in particular if you have enough taxa, say > 100.

What I say here is just my intuition about it, wi8thout any hard
evidence. In your case I'd suggest that you run the Bayesian analysis
with the models returned by Modeltest, and then simply once more by
using GTR+G for all partitions in the Bayesian as well as ML analysis
and then try to quantify the differences, the interesting question being
if the tree topologies will be different.

Hope this helps a bit,

Alexis

> As far as I understood I can only define different GTR models as a command
> in combination with my partition file (e.g. -m GTRGAMMAI -q partition). The
> GTR model would thus be assumed for all the partitions however. Is there
> any way to include the different models in the partition text file, such as
> described for the different protein models?
>
> I would be very grateful for advice!
>
> Cheers Hannah
>
>

--
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

[Hannah Miller]

Dear Alexis,

thanks a lot for your reply.
I indeed saw a slight difference when using only the GTR+G model versus the models suggested by Modeltest in the bayesian tree topology.
I will try the GTR+G model in RaxML as suggested and compare the differences. I don`t expect major changes in topology, but good to know that indeed the specific model definition per gene is not possible with RaxML.

Thank you!
Hannah

[Alexandros Stamatakis]

:-)

Alexis

[Lee Van]

Dear Alexis and Hannah,

I am also stuck in this problem recently. After 4 years from that time, is it possible to assign different models now using RAxML-HPC BlackBox on CIPRES?

Regards,

Lee

在 2014年6月4日星期三 UTC+8下午6:06:36，Hannah Miller写道：

[Grimm]

Hi Lee,

No. RAxML still has only the one basic model for nucleotides.

Situation hasn't changed, what Alexi pointed out still holds.

There's is also no point in implementing more restricted models, according to various experiences with various data sets over the years. If your nucleotide partitions follow a certain model, RAxML has a certain chance to approximate this.

You can check this in the RAxML_info files generated during the run. If you use different partitions, and they may prefer different models, you will see substantial differences in the optimised mutation probabilites of the individual GTR models for each partition.

I personally had cases where some partitions converged to e.g. a HKY model, while others ended up with a full parameter GTR model (all substition probabilities different).

Cheers, Guido

[Alexey Kozlov]

Hi Lee,

it is possible with the new RAxML-NG BlackBox available here:

https://raxml-ng.vital-it.ch/#/

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]

This is possible with RAxML-NG now:

https://github.com/amkozlov/raxml-ng

a web-server offering RAxML-NG is available here:

https://raxml-ng.vital-it.ch/#/

alexis

> --
> 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 (Alexis) Stamatakis

Research Group Leader, Heidelberg Institute for Theoretical Studies
Full Professor, Dept. of Informatics, Karlsruhe Institute of Technology

www.exelixis-lab.org