[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]

Re: Evolution programs (was: Reevolving bones)



At 03:25 AM 5/7/97 -0400, Dinogeorge@aol.com wrote:
>This is exactly what I had in mind. Have the evolution program generate an
>"ecosystem" of virtual dinosaurs, hand the dinosaurs over to the
>systematists, then see whether the trees the systematists come up with match
>the trees that the computers constructed when they generated the virtual
>dinosaurs. Even more interesting: Set the parameters of the programs to
>generate trees with no reversals, lots of reversals, no homoplasies, lots of
>homoplasies, strict parsimony, no parsimony, and so forth, and see whether
>the methodology can consistently pick them out. Or see just what it does do
>with such trees. This could go any number of ways...
>
Actually, this is very similar to the method of establishing likelihood
ratios for DNA phylogenies that was described in the recent article.  The
method involved estimating the parameters of the DNA evolution of the group
from the data set, and using that to parameterize a set of Monte Carlo
simulations of the evolution of the group, and deriving the likelihood of a
given set of tree statistics from the set of results of the simulations.


The main limitation in applying this to character based phylogenies is the
problem of parameterization.  For DNA and protein sequence phylogenies
there are clear, and estimatable, parameters, such as net mutation rates,
the variance and covariance of mutation rates, and so on.  (The article,
which I do not have here, lists a set of parameters they used for their
study).

I just cannot quite see how to do an equivalent parameterization of the
simulations from a character matrix.  What features of the distribution of
characters would display the fundamental dynamics of the evolution of those
characters?

--------------
May the peace of God be with you.         sarima@ix.netcom.com
                                          sfriesen@netlock.com