> To find the optimal mutation rate is a tricky business. Too low and
> there is no evolution at all, too high and no useful information can
> accumulate. Furthermore, also in biological reality the mutation rate is
> not fixed and given, but is itself subject to evolution through
> mutations in DNA repair, proofreading and antioxidant genes. So it seems
> only natural to me if the mutation rate could somehow be part of each
> individual and be up or downregulated via mutations. Hopefully, this
> would lead to an optimal mutation rate.
>
> Is something like this possible (with GP) ?   Has this been discussed
> before ?
Great. Topic has been my intrest for long time. You can implement that
various ways. It definetely is possible.
I am glad to see somebody else is intrested topic. Also biological facts
that cross-overing cut points are not fixed or statistically homogenous.
Instead cross-overing cutting points vary in biology.

Following is not exactly GP but you can implement such simple thing very
simple way by adding some structural data
for customized genome. Like for example mutation rate. Then you just use it
in mutation function.

"my earlier posting for comp.ai.genetic"

"- Structural alphabet

Functional information in genome affects individual behaviour.

Structural information in genome affects genetic operations.

Many EA:s use only functional information. Biological genome contains lot of
structural information.
Simplest example of structural information are chromosomes. I have about 20
other examples what I see as structural information.

It is possible that two genes can be functionally equivalent but
structurally different. This means that two alleles have same effect for
individual surviving possibilities in current enviroment but alleles can
have different properties in evolutionary lineages.

It is also possible that two individuals are partially separate species. It
those individuals have so different chromosomes A1 and A2 that those can't
crossover (only swap). Makes species consept more concrete, or what you
think? So in this context womans and mans are partially different "species"
because having separate x and y chromosomes. I see that that also other
cromosomes can go different paths that cross-overing is not anymore
possible. (or like EA:s in general you can see it
also purely propablistic issue). This same thing goes also gene level that
genes can be so different that those only swap if genes are functionally
similar but structurally different.

This are my basic prinsibles for structural alphabet where I investigate how
different structural properties in alphabet work in evolutionary algorithm
context. I have defined three properties for alphabet "stability",
"connectivity" and "functionality". So "codon" can mutate so that only it
stability property changes for example or gene can get new hot-point.

I only know some earlier studies where some properties but not many are
investigated. Earlier studies cover hot-points, varying mutation rates. This
model fills my views for lineage selection descriped earlier in this group.
My references are mainly for papers created 1994-2000, so I am intrested
later studies considering ea investigations of alphabet structural
properties (like I defined it, definition is my own). I get zero references
when I asked lineage pairing studies, so papers studying structural
properties could be great suprise, (GP/GA) both ok, or still also intrested
papers investigating lineage pairings."

Hopefully previous was not too of topic. I am just ecj user.
Author of http://sourceforge.net/projects/narugo/.

t. Harri