Re: [Swarm-Modelling] population genetics

[Alex Lancaster]

>>>>> "SC" == Scott Christley  writes:

SC> Greetings, Is anybody aware of research with an agent-based model
SC> (or other modeling approaches) that simulates evolution
SC> (recombination/ mutation) on actual genomic data?  I'm aware that
SC> genetic algorithms often use an artificial genome with attributes
SC> encoded in artificial genes, but I am looking for anybody who is
SC> using actual nucleotide sequence.  For example, a population
SC> genetics study may have sequenced a number of alleles for an
SC> organism, then use those alleles (and possible allele frequencies)
SC> to simulate and predict how environmental changes could affect the
SC> population diversity.  I am able to find empirical studies which
SC> record allozyme diversity across species, spatially, etc. but I
SC> have not found anybody doing simulations with the actual genomic
SC> data.

Hi Scott,

In population genetics the focus has moved from forward simulation
(like the Wright-Fisher or Moran models) to coalescent theory
approaches.  Many coalescent theory approaches start with real genomic
data (i.e nucleotide or amino acid data) and can be used to estimate
various parameters about the populations typically to predict age of
alleles or other parameters of a population such as effective
population size.  

A good introduction to coalescent theory is John Wakeley's new book,
the first few introductory chapters are available for free online:

http://www.roberts-publishers.com/wakeley/

Having said that, coalescent approaches have some important
limitations in the situations (at least in the basic coalescent
approach) that they apply to, for example selection and recombination
are hard to include explicitly.

I have heard of (but haven't tested) CoaSim which is an environment
for simulating genetic data using coalescent models: 

http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1274299

In some of these cases, forward simulation approaches are still useful
or necessary and chapter 3 of Wakeley book introduces those techniques
as well.  In terms of implementing those models, there aren't that
many general purpose environments, but there are plenty of sample
implementations in various languages around.  Use the new Google Code
search and search for Wright-Fisher (or Fisher-Wright).

There has been some work trying to bridge the gap between research
using A-life/GA/evolutionary computation approaches and those in more
traditional population genetics, most notably Richard Watson (recently
a postdoc with John Wakeley's group) had a workshop at A-Life 9 on
just that (which I, unfortunately could not attend):

http://demo.cs.brandeis.edu/alife9/watsonTut.htm

Not sure if that was written up as a paper or not, but I'm sure you
could contact him.

What exactly do you need to simulate?  i.e. what is your input data,
do you need to keep track of genotypes as well as allele frequencies?
what are you trying to measure?  If you trying to estimate parameters
from data then some prepackaged things may work for you, but if you
need to introduce some dynamics in each step of the simulation, then
you probably need to build your own model (perhaps using some existing
code/algorithms but with additional features that you would probably
need to write from scratch).  You could do it in Swarm, but it may be
an overkill if your agents are simply genotypes rather than being
embedded in organisms with their own dynamics.

Hope this helps,

Alex