Q: Difference of tree search between GTRGAMMA and GTRCAT

[Akifumi]

Dear all,

I found that RAxML shows slightly better performance under GTRCAT than
under GTRGAMMA even if the values of initial rearrangement setting are
equal.
Are there the differences of tree search between GTRGAMMA and GTRCAT?

Best regards,

[Alexis]

Dear Akifumi,

I am not sure what you mean by better performance. Is this with
respect to the final GAMMA-based scores of searches
under CAT and GAMMA respectively or do you refer to inference times.

Generally, CAT may perform better than GAMMA both with respect to
inference times (that should always be the case),
but may also produce trees with better scores.

In the latter case, CAT may yield trees with better scores under GAMMA
because the model is slightly different and hence, even when
using the same starting tree it may chose a different search path
through tree space.

For details please have a look at this paper:

http://wwwkramer.in.tum.de/exelixis/pubs/HICOMB2006.pdf

Alexis

[Akifumi]

Dear Alexis,

Thank you for your reply.

When I compare the resulting tree topologies among shotgunned tree
search with many starting trees under GTRGAMMA, one-time heuristic
search under GTRGAMMA, and one-time heuristic search under GTRCAT
(same starting tree was used under GTRGAMMA and GTRCAT), the resulting
tree topology of GTRCAT was closer to that of shotgunned search than
that of GTRGAMMA in many cases.
The GAMMA-based lnL of resulting tree topology of GTRCAT is larger
than that of GTRGAMMA in several cases in your paper.
However, I cannot understand yet why GTRCAT provides better results
than GTRGAMMA even if starting tree and search algorithm are the same.

Best regards,

Akifumi

[Alexis]

Dear Akifumi,

Well it's just a different way (and different model) for capturing
rate heterogeneity among sites.
Trees inferred under GTRCAT may thus be better than trees inferred
under GAMMA on your specific dataset,
although this does not need to always be the case. In general GAMMA-
based scores of trees inferred under CAT and GAMMA
should be pretty comparable in general.

Does that help?

Alexis

[Akifumi]

Dear Alexis,

I've been convinced.

Thank you!

Akifumi

[Zhang Rong]

Dear Alexis, 

Are the GAMMA, or GTRGAMMA and CAT models site-heterogeneous models in RAxML?
Thanks.

Rong
在 2011年8月7日星期日 UTC+8下午1:55:59，Alexis写道：

[Alexandros Stamatakis]

yes,

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>.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/raxml/42622eca-30bd-4c11-ab2a-acd45d637084%40googlegroups.com
> <https://groups.google.com/d/msgid/raxml/42622eca-30bd-4c11-ab2a-acd45d637084%40googlegroups.com?utm_medium=email&utm_source=footer>.

--
Alexandros (Alexis) Stamatakis

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

www.exelixis-lab.org

[Zhang Rong]

Thank you very much. In some papers, such as "site-heterogeneous models (CAT + Γ and CATGTR + Γ) perform slightly better than site-homogeneous models (LG + Γ and GTR + Γ)" (Ref: Gouy et al., 2015, Rooting the tree of life: the phylogenetic jury is still out).  This could be performed in Phylobayes. Could you give me some suggestions to understand these differences of GTRGAMMA in RAxML and Phylobayes? Is it also a site-heterogeneous model in Phylobayes?

Thanks again.

Rong

在 2019年7月31日星期三 UTC+8下午5:12:51，Alexis写道：

> an email to ra...@googlegroups.com
> <mailto:ra...@googlegroups.com>.

[Alexandros Stamatakis]

The naming was unfortunate, the CAT model in RAxML is completely
different from the CAT model in PhyloBayes, in RAxML the CAT model is
just a computationally cheap workaround for modeling rate heterogeneity.

You can find the paper describing it here:

http://www.hicomb.org/HiCOMB2008/papers/HICOMB2006-05.pdf

Alexis

On 31.07.19 11:30, Zhang Rong wrote:
> Thank you very much. In some papers, such as "site-heterogeneous models

> (CAT + /Γ/ and CATGTR + /Γ/) perform slightly better than
> site-homogeneous models (LG + /Γ/ and GTR + /Γ/)" (Ref: Gouy et al.,

> > an email to ra...@googlegroups.com <javascript:>
> > <mailto:ra...@googlegroups.com <javascript:>>.

> > To view this discussion on the web visit
> >
> https://groups.google.com/d/msgid/raxml/42622eca-30bd-4c11-ab2a-acd45d637084%40googlegroups.com
> <https://groups.google.com/d/msgid/raxml/42622eca-30bd-4c11-ab2a-acd45d637084%40googlegroups.com>
>
> >

> <https://groups.google.com/d/msgid/raxml/42622eca-30bd-4c11-ab2a-acd45d637084%40googlegroups.com?utm_medium=email&utm_source=footer

> <https://groups.google.com/d/msgid/raxml/42622eca-30bd-4c11-ab2a-acd45d637084%40googlegroups.com?utm_medium=email&utm_source=footer>>.
>
>
> --
> Alexandros (Alexis) Stamatakis
>
> Research Group Leader, Heidelberg Institute for Theoretical Studies
> Full Professor, Dept. of Informatics, Karlsruhe Institute of Technology
>

> www.exelixis-lab.org <http://www.exelixis-lab.org>

>
> --
> 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>.

> To view this discussion on the web visit

> https://groups.google.com/d/msgid/raxml/6fe485d3-e5e4-4746-b522-a31f980bcc9e%40googlegroups.com
> <https://groups.google.com/d/msgid/raxml/6fe485d3-e5e4-4746-b522-a31f980bcc9e%40googlegroups.com?utm_medium=email&utm_source=footer>.

[Zhang Rong]

OK. Thanks Alexis.

在 2019年7月31日星期三 UTC+8下午7:04:05，Alexis写道：

> an email to ra...@googlegroups.com
> <mailto:ra...@googlegroups.com>.