is exabayes dirichlet prior analogous to the phylobayes implementation (aka CAT)?

[Xavi Grau]

Hi,

I've seen that exabayes implements the possibility of specifying dirichlet priors for GTR substitution rates ("Reversible Matrix Prior" section in the manual) using the revMatPr dirichlet(1,...1) parameter, both for nucleotide and aminoacid data.

I'd like to know if this process is similar to so-called CAT-like models of phylobayes. Taken from the phylobayes v1.5 manual:

PhyloBayes uses Dirichlet processes for modeling sites-specific profiles (Lartillot and

Philippe, 2004). Each site is thus given a frequency vector profile over the 20 amino-acids

or the 4 bases. These are combined with a globally defined set of exchange rates, so as to

yield site-specific substitution processes. The global exchange rates can be fixed to uniform

values (the CAT-Poisson, or more simply CAT, settings), to empirical estimates (e.g. JTT,

WAG or LG) or inferred from the data (CAT-GTR settings).

My doubt arises from the fact that exabayes has so many similarities with RAxML, and this software doesn't implement the CAT-like models of phylobayes, is that correct? Taken from the Raxml v8.2 manual:

The CAT model of rate heterogeneity

The name of this model has caused a lot of confusion because there is a CAT model also

implemented in PhyloBayes (see http://megasun.bch.umontreal.ca/People/lartillot/www/index.htm)

Unfortunately, I was not aware of this when I introduced the CAT model in RAxML, the CAT model in

RAxML is something completely different! However, I decided not to change the name for

backward compatibility such that new RAxML version keep working with old wrapper scripts.

 Thanks a lot in advance!

Xavi

[Andre Aberer]

Hi Xavi,

I agree, these are quite a few terms that sound similar, which is
confusing, but they mean totally different things.

The ExaBayes model is identical to what MrBayes offers: you specify a
prior belief about the rates in the matrix (and if you set all values to
1, this belief is effectively uniformly distributed).

Dirichlet processes are a completely different thing and ExaBayes cannot
do Lartillot's CAT model to model rate heterogeneity. It does not offer
Stamatakis CAT approximation (now called PSR) either. Only the Gamma
model is supported.

--
Best regards,
Andre

[Xavi Grau]

Hi Andre,

Thanks for the clarification. These nomenclature issues are indeed confusing to us, the math-agnostics.

Best,

Xavi

El dia 01/06/2016 8:51 a. m., "Andre Aberer" <an...@aberer.io> va escriure:

--
You received this message because you are subscribed to a topic in the Google Groups "ExaBayes" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/exabayes/F29HovbF8Ks/unsubscribe.
To unsubscribe from this group and all its topics, send an email to exabayes+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.