Prev: Constructive evolution (overview)
Next: RNA evolution


Constructive evolution


In our assessment of the role of space and the effects of interlocking evolutionary and ecological dynamics, our main approach has been to take classical population dynamic models and augment them with space and ongoing phenotypic mutations. In this way we have included local interactions and have allowed for mesoscale patterns to arise. Moreover, ongoing mutations contrast with the more classical approach of changing parameters. In this setting we have studied the persistence and invasion dynamics of different phenotypes and thus optimization of parameters in the system. In this section we deviate from this approach and take a different perspective, namely one focusing on constructive evolution.
external image baby_elephant.jpg
In the 70's, Maynard Smith emphasized the fact that although we know about fitness and mutations, and thus natural selection, that doesn't tell us about how evolution constructs new things. Maynard Smith therefore suggested that every evolutionary biologist should go to the zoo once a year and stand in front of an elephant and say:
"Elephant, I believe you came about by random mutations!"

In the following our goal is therefore to try and find a way to extend our modeling in order to allow for constructive processes to occur by mutation and selection. To guide us in this enterprise we make use of several useful observations:
  • the number of genotypes is much greater than the number of individuals. Individuals are therefore always a subset of genotype space (vs quasi-species equations where all mutants are present).
  • there is a complex genotype-phenotype mapping. This mapping is not very direct (without which you don't get nice Mendelian inheritance), but should be very important, and mutations occur on genotypes but selection occurs on phenotypes.

Genetic algorithms

Such an approach requires the view of Darwinian selection as an efficient design, or optimization, tool. To this end, John Holland (1975) came up with genetic algorithms, or evolutionary computation, which incorporate:
  • a population of (coded) structures, or solutions, or cases
  • some fitness criterion
  • replication, or reproduction
  • decay, or death
and very importantly
  • mutational operators: not just point mutations but also duplications, deletions, rearrangements.
(In fact Holland (1975) was one of the first to really think about the importance of such processes.)

Such genetic algorithms are used for various systems with complex counterintuitive behaviour in order to design an optimized system such as:
(Note: even nanotechnology requires evolutionary optimization, since intelligent design apparently doesn't work! Therefore it is even more paradoxical for nanotechnology expert Cees Dekker to be such a proponent of intelligent design!)

Genetic algorithms therefore lead to a shift in methodology. Before we were concerned with changing parameters and saw that according to reproduction rates the direction of evolution could change. The focus was on what evolves. Now we set an external fitness criterion (goal) and things get fitness when they are closer to that goal. In this way we allow the system to discover how to achieve that goal. The focus is on how things evolve. With this shift in methodology we first study RNA evolution.

Next: RNA evolution


References

Holland HH (1975) Adaptation in Natural and artificial systems. Univ of Michigan Press 183 pp. (reprinted 1992 MIT Press)




So far we have looked at:
- classic population dynamic models
- added space (local interaction and mesoscale patterns)
- added ongoing phenotypic mutations (vs just changing parameters)
- and looked at: who persists/invades and parameters of optimization

Now we take a different perspective: Constructive evolution
"Elephant, I believe you came about by random mutations!" Maynard Smith (197?).

- we know about fitness and mutations etc
- but that doesn't tell us about HOW you construct NEW things in evolution

Or goal is therefore to try to find a way to extend our modelling in order to allow for "constructive" processes to occur by mutation and selection. In this there are several important observations:
1 - num genotypes >>> num of individuals: i.e. always a subset of genotype space (vs quasi species equations where all mutants are present)
2 - genotype phenotype mapping, where mutations occur on genotype but selection occurs on phenotype. This mapping should be very important. In most cases it is probably not very direct (without which you don't get nice Mendelian inheritance).



Such an approach ask for the view of Darwinian selection as an efficient design (optimization) tool.
Holland: came up wtih genertic algorithms / evolutionary computation
- population of (coded) structures / solutions / 'cases'
- fitness criterion
- reproduction and decay
and very important:
- mutational operators: not just point mutations but also duplicaton, rearrangements, deletions (Holland was the first to really think about the importance of such processes).

Such genetic algorithms are used for various systems with complex unintuitive behaviour in order to design an optimized system.
examples:
- computer networks / job selection
- robotic control / body design
- nanotechnology (where intelligent design apparently doesn't work! Kees Dekker eat your heart out)
- in vitro evolution of ribozymes (RNA world)

So in terms of methodology:
Before: change parameters and according to reproduction rates see direction of evolution can change etc: WHAT EVOLVES
Now: set exteral fitness criterion (goal) and things get fitness when they are closer to that goal and allow system to discover how to achieve that goal: HOW EVOLVES