Re: [Swarm-Modelling] population genetics

[Alex Lancaster]

>>>>> "MD" == Marcus G Daniels  writes:

MD> Quoting Alex Lancaster <address@hidden>:
>> 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.

MD> Even simple nucleotide substitution models like REV, require
MD> ENORMOUS computational resources to infer an evolutionary history.
MD> And to do it on more than 500 sequences is pretty much impossible
MD> without a supercomputer.

True, but it all depends on whether you are trying to infer an
evolutionary history, which is a backwards-in-time huge combinatorial
problem as you say, which is more of a phylogenetics problem.  Doing a
what-if qualitative forward-in-time population genetics simulation is
potentially more tractable.  It seems that Scott's problem is more the
latter.  Although I agree with your implicit suggestion that Swarm
would probably be too heavy for such a model. 

It also depends on how many loci you are interested in simulating, and
how many alleles at each loci and whether you want to include
recombination.  Of course, the problem with forward-in-time
simulations is that you need many runs with different parameters
and/or random number seeds to fully explore the parameter space.  

MD> Substituting more complex models (e.g. with subpopulations having
MD> different environmental conditions / selection pressures) makes
MD> these costs all the worse!

If you have a pretty good idea of the parameters involved and your
number of alleles/loci isn't too large, then a Wright-Fisher approach
is feasible (with the addition of some selection and/or individual
dynamics) and there are some simplifications you can make if you have
fairly large population sizes.  Moran models are also fully exact and
can deal with stochastic issues like small population size but are
only well developed for haploid organisms, AFAIK.

Alex