parsimony/biology (was: Striking a balance...)

[Pierre Deleporte]

(I renamed this part of the thread so that Peter Hovenkamp can delete it
more easily).

At 12:42 21/02/01 Tom Di Benedetto wrote:

>>Pierre Deleporte wrote:
>>...under what biological criteria do we apply the "parsimony criterion"? 
>
>This use of the parsimony criterion makes no reference to biology or
>evolutionary theory per se; it is used as a logical tool in the process of
>extracting patterns from evidence that is not entirely consistent.

Well, I think this is exactly what you logically cannot pretend to do. And
Kirk Fitzhugh stated it well in his last message.
A colleague suggested me a reformulation for "biological criteria" (may be
confused with "optimality criterion"). I meant : under which biological
explanatory law do you apply parsimony?

"Descent with modification" in itself is not sufficient to distinguish
between different phylogenetic methods.

I believed that my allusion to "Theorem 5" of Tuffley and Steel (1997) was
sufficient to put light on this:
"Theorem 5: Maximum parsimony and maximum likelihood with no common
mechanism are equivalent in the sense that both choose the same tree or
trees."

ML corresponds to "some common mechanism", MP corresponds to "no common
mechanism" (local optimization of homology).

The original data matrix (primary homology) is the same for ML and MP (and
for a phenetic/molecular clock approach too).

The particular vision of the evolutionary process in standard MP ("no
common mechanism") is definitely NOT included inside the mere definition of
primary homology (data matrix).

I agree that standard MP globally optimizes evolutionary scenarios for
characters, but it does this SOME WAY ("no common mechanism" etc...). Other
ways are imaginable.

If you agree with these points (and you agreed with some of them in your
previous messages), how can you pretend that the choice of parsimony as
applied in MP has no biological implication?

Alternative proposition (silly proposition given the current most popular
vision of the evolutionary process): I could parsimoniously maximise
homoplasy instead of parsimoniously optimize character state contiguity.
The explanatory law would be : horizontal transfer of characters is the
overwhelming evolutionary mechanism, and primary homology is systematically
misleading.

But some models in ML implement some notion of decreased confidence in the
informativeness of primary homology in some circumstances... long branches
etc... and they do this with a notion of "common mechanism" throughout the
tree.

Given the matrix of putatively homologous characters, applying parsimony
the standard way for optimizing character state contiguity with "no common
mechanism" IS a biological decision, not a mere general application of
Occam's 
razor "like that".

Occam's razor may be used to cut off a lot of different things, under
different logics.

Pierre


Pierre Deleporte
CNRS UMR 6552 - Station Biologique de Paimpont
F-35380 Paimpont   FRANCE
Téléphone : 02 99 61 81 66
Télécopie : 02 99 61 81 88