Molecular taxonomy: on way out?

[pierre deleporte]

A 13:13 18/07/2005 -0400, John Grehan wrote :

> > Plesiomorphies are IGNORED by the algorithms...
>
>Not if they are treated as apomorpies.

But they are NOT, John, they are not!

Of course, contemporaneous phylogeneticians accept the outgroup criterion 
for rooting the trees, hence optimizing the plesiomorphy-apomorphy 
polarization for most of the data, and this is done by the algorithm during 
the analysis. Everybody knows that (a few hours training is sufficient).

But yo seem to reject the outgroup criterion, and you pretend to polarize 
every character a priori, and put only what you consider reliable 
apomorphies into the data matrix.
If I understand well, your criterion is that a character(-state) never 
present in ougroups is a reliable apomorphy (all what I could pick out of 
your different messages on the topic, including your practice in the Pongo 
debate).
Well, if this is correct, you are implementing yourself the outgroup 
criterion you reject in other peoples' approaches! Don't reject your own 
practice when nicely performed by algorithms.

Also, it seems that you don't put into the data matrix characters(-states) 
which happen to be present in some outgroups.
Well, if this is correct,  you are implementing a clique analysis some way! 
(I doubt that Nelson would approve that...). i.e. you are cool-bloodly 
killing non-completely congruent characters, hence denying the possibility 
that they may be convergent at some high level in the phylogeny, but 
informative at a lower level. Have you any justification for this?

You wrote :

 >The result for molecular genetics is that characters
 >that are really primitive may be treated as being derived and there
 >seems to be no independent test of the choice because the sequences
 >cannot be shown to be primitive or derived in the first place sequence
 >by sequence (at least that is my understanding... ).

It would be interesting that you explain how morphological systems (the 
rough equivalent of sequences in phylogenenetic analyses) can be 'shown to 
be primitive or derived in the first place'. What is your method? Without 
implementing the outgroup criterion, of course, because you deny it for 
molecules, hence you should'nt use it for morphology. Sorry, I remember 
now, in fact you implement the outgroup criterion.
Or do you rather look carefully at characters, in the blank of the eyes, 
and get some intuitive illumination?
Or do you implement some model of morphological character evolution? Which 
one and why? And if so, why reject molecular model-based analyses?

Molecular-clock approaches (tentative datation of phylogenetic divergence) 
can use phenetic algorithms, which is all right under the model of regular 
divergence. This is debatable, of course; but NOT all molecular analyses 
are molecular-clock-like.

Another commentator correctly pointed that characters cannot be 'phenetic' 
(shear nonsense of course, but this has been repeatedly explained to you on 
this list since years now).

 > ...and I am interested to be shown to be wrong

Well, just done once more, in my view (or prove ME wrong, for once, instead 
of merely repeating your point).

addendum: your last post sheds more light on your very unique way of 
implementing 'cladistics': you put only (putative?) synapomorphies into the 
data matrix. You reject a priori all characters not included in this clique.
And you implement the outgroup criterion, but refuse that the algorithm 
does it for you.
And you don't say how you deal with contradictions between your different 
putative synapomorphies, possibly pointing to different and mutally 
incompatible clades: what is your optimality criterion, and why not let the 
algorityhm implement it for you?
Unless you have no clear compatibility criterion, and you go on 'pruning' 
the data set down to a completely compatible clique in your brain before 
the 'analysis' (already done in fact!)?
Not to mention your possible criteria for a priori ordering all the 
multistate characters: any rule? (don't say outgroup, it cannot work beyond 
two states).

A great difficulty in the debate is your quite unique, very personal 
version of 'cladistic analysis'. I really know of nobody performing 
phylogeny inference that way (and certainly not Nelson, one of your pet 
references: he doesn't and never did as far as I know, certainly not 
character assassination!).

Pierre


Pierre Deleporte
CNRS UMR 6552 - Station Biologique de Paimpont
F-35380 Paimpont   FRANCE
Téléphone : 02 99 61 81 63
Télécopie : 02 99 61 81 88