[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]
The Neoaves paper again (and JP1)
(If you're just interested in Jurassic Park, scroll to the bottom.)
A kind soul has sent me the supplementary information.
- If analyzed alone, each gene contains rather little phylogenetic signal.
Clearly, piling up more data helps.
- Taxon sampling is crucial, and not just in the ingroup. If the paleognaths
are omitted (so that Neoaves is rooted by the galliform and the anseriform
alone), the root of Neoaves is suddenly inside Metaves (giving 3 separate
successive outgroups to Coronaves: the Caprimulgiformes-Apodiformes
assemblage, the rest of Metaves except the hoatzin [completely resolved at
95 or more % posterior probability], and then the hoatzin which is the
sister-group to the rest of Neoaves).
- The model of evolution under which each gene was analyzed (you can have
different models for different parts of the data matrix in the same
analysis) was chosen by the program Modeltest. Well, what else. Modeltest
allows you to estimate the model and its parameters from the data, as
opposed to having to pull both out of thin air. The only problem is the way
Modeltest does that. It constructs (Debruyne & Tassy 2004 for Modeltest
3.06; pers. obs. for Modeltest 3.7) a neighbor-joining tree using the
simplest model (Jukes-Cantor -- basically, no model at all: equal
frequencies for all 4 bases, equal probabilities for all mutations). It does
not show you that tree. Then it tests 56 models and chooses the two that fit
the tree best according to 2 different criteria, and based on which
criterion you prefer, you use one of those models. But -- hello --
neighbor-joining is PHENETICS. And phenetics is EVIL. It is not
phylogenetics. It is just a measure of allover similarity. It HAPPENS to
give the same topology as the science of phylogenetics does _if_ there is
little enough homoplasy in the data -- but that is precisely what we're
trying to test!!! The only reason I can imagine for why Modeltest uses a
neighbor-joining tree calculated using the Jukes-Cantor model is that a)
this is the fastest method to get something tree-shaped out of any dataset,
and b) there are plenty of molecular biologists who have simply not
understood the difference between phylogenetics and phenetics (I've tried to
discuss with one).
- The calibration points for the dating are all internal, and most of them
are from Mayr papers, so I'm not able to criticize them. The only external
calibration point is the Galliformes-Anseriformes divergence: the earliest
stem-galliforms known are from the early Eocene, and this is the external
calibration point. *Vegavis* was apparently published too late.
- The use of two different methods for estimating the divergence dates is
very interesting. So, next time you see PL (penalized likelihood) being
used, consider all dates it gives too old by (in this case) on average 21
million years. Probably that's because it biases against too large rate
changes between ancestors and descendants, rather than between sister-groups
as the other method used in that paper does. That said, the other method
fails to predict the discoveries of *Vegavis* and even *Waimanu*.
Régis Debruyne & Pascal Tassy: Vers une phylogénétique non systématique ?,
Biosystema 22, 25 -- 34 (2004) (French with English abstract -- that's
right, French, and in a relatively obscure journal. I wouldn't have found it
on my own.)
Now, on the issue of the book Jurassic Park being brilliant... let's put it
this way: it's better than the book Lost World.
Seriously... Crichton seems to have thought in earnest that *Ornitholestes*
and everything similar were actual real true lizards. Complete with shedding
the entire skin at once (only squamates do that, and if Crichton would do a
little thinking, he'd find out that he knows this full well!!!) and some
fighting behavior that even most lizards probably don't have. No, the
science in that book is not a lot better than that of State of Fear, and the
plot could benefit from a better, no, any amount of understanding of chaos
theory. Look, it's just a Greek or Shakespearean tragedy: everyone who has
in any way loaded guilt upon their shoulders dies.
Now, if it were a Kishon tragedy...