----- Forwarded message from
morphmet_modera...@morphometrics.org -----
Date: Sun, 22 Dec 2013
00:20:49 -0800
From: morphmet_modera...@morphometrics.org
Reply-To: morphmet_modera...@morphometrics.org
Subject: Re: Discussion
on GM and phylogeny
To: morphmet@morphometrics.org
----- Forwarded message from Santiago Catalano
<sacatal...@gmail.com> -----
Date: Thu, 19 Dec 2013 15:01:30
-0500
From: Santiago Catalano <sacatal...@gmail.com>
Reply-To: Santiago Catalano <sacatal...@gmail.com>
Subject: Re:
Discussion on GM and phylogeny
To:
morphmet@morphometrics.org
I would like to make some comments about the messages from Aki Watanabe and Joe Felsenstein:
1-. About the independent treatment of individual landmark:
Our approach (Catalano et al. 2010; Goloboff & Catalano 2011; Catalano & Goloboff 2012) shares this with those analyses that use Procrustes Distances (PD) to quantify differences in shape. Some researchers tend to think that PD are describing differences related to changes at the “configuration level”, while in fact the quantification of shape differences by PD is also completely defined by a sum of individual landmark differences, being affected by landmark covariation and landmark choice in the same way than our approach.
2-. We were aware about the covariation problem and we discussed this issue in our three papers. Taking into account the covariation among landmarks would be clearly an improvement over our approach. However, I have doubts about the empirical utility of a method that intends to take into account covariation among landmarks. Based on the analysis of several datasets, I often see that covariation among landmarks strongly varies along the tree. Hence, the improvement should come from a method that can recognize covariation among landmarks that changes along branches of the tree.
3-. Our method (just like methods that use PD) is affected by covariation. Were this an argument to just discard our method, then one should also discard probabilistic models because, in the context of landmark analysis, they are affected by making unrealistic assumptions. Defending those probabilistic models on the ground that “it is the best that can be done” would be a clear double standard.
4-. The use of probabilistic approaches is based on a philosophical mandate as much as not using them. Although obviously very useful in several areas of research, statistical inference is not the only way to do science. In particular, I see no improvement in using a probabilistic approach when the models considered are far from realistic.
5-. Significant improvements in the use of landmark data in phylogenetics may come from incorporating models of shape change (models in general not in the probabilistic sense), that can better quantify the differences in shape. For instance incorporating transformations that affect configurations as a whole -these models obviously are also considering a kind of covariation. And this improvement can be done in a probabilistic context as much as in a parsimony context.
Santiago
Santiago
----- Forwarded message from Joe Felsenstein <j...@gs.washington.edu> -----
Date: Wed, 4 Dec 2013 10:03:10 -0500
From: Joe Felsenstein <j...@gs.washington.edu>
Reply-To: Joe Felsenstein <j...@gs.washington.edu>
Subject: Re: Discussion on GM and phylogeny
To: morphmet@morphometrics.org
In reply to Shobnom Ferdous, who asked:
>> Why we cant incorporate a morphological
>> phylogeny or Geometric morphometrics analysis,
>> Molecular phylogeny all together to get a
>> phylogeny for a particular group, rather we use
>> a molecular phylogeny and overlay that with GM
>> work? I have read the related papers but some
>> discussion is much appreciated.
Aki Watanabe said:
> Geometric morphometrics can be incorporated
> as characters in a phylogenetic analysis (see
> Goloboff and Catalano papers in the journal
> Cladistics for parsimony analysis of aligned GM
> data), but I don't think it has been implemented
> in maximum likelihood or Bayesian analysis
> before perhaps because choosing probabilistic
> and prior distribution models for evolution of
> landmark data is a tricky issue.
>
> That said, I personally prefer NOT to use GM
> data as phylogenetic data because of few
> reasons: (1) phylogenetic characters need to be
> independent of each other, but landmarks that
> are close together will likely not be
> independent; (2) it is suspicious that
> individual landmarks appropriate evolutionary
> units, especially whether "geometrically or
> functionally" homologous landmarks are
> "evolutionarily" homologous; and (3) if x, y,
> (and z) coordinates of each landmark are taken
> as separate characters, then the result will be
> dependent on the orientation of the specimens
> after alignment (Goloboff and Catalano's method
> treats each landmark as character so this
> problem doesn't apply to their method).
A few comments: People have been using landmark
coordinates (appropriately superposed) in
phylogenetic comparative methods since at least
2002. Those methods are based on Brownian
Motion models of correlated quantitative characters,
using likelihood inference. Bayesian inference
is possible for that too. Of course, the
characters covary -- in fact, inferring those
covariances is the whole point in that case.
For GM coordinates or in fact for any quantitative
traits, it is not appropriate to use them for
inferring phylogenies unless you have some way
of dealing with the covariances. Assuming that
they are i.i.d. is bad statistical practice, as
Watanabe sees.
This applies to Goloboff and Catalano's method
if that method is considered to be making a
statistical inference. But if G&C intend some
other philosophical framework then all that
need be noted is that their method might be
philosophically mandated ... but that does not
settle what are its statistical properties.
Joe
----
Joe Felsenstein j...@gs.washington.edu
Department of Genome Sciences and Department of Biology,
University of Washington, Box 355065, Seattle, WA 98195-5065 USA
----- End forwarded message -----
Dr. Santiago A. Catalano
INSUE - UNT
SM Tucumán - Tucumán
Argentina
----- End forwarded message
-----
----- End forwarded
message -----
