Richard A. Watson and Daniel M. Weinreich
Organismic and Evolutionary Biology, Harvard University.
Synopsis: A tour of basic evolutionary genetics
for evolutionary computation practitioners.
We provide a description of some of the basic models and
assumptions used in evolutionary genetics and population genetics as
practiced in evolutionary biology. We contrast these models and
assumptions, appropriate for life as it is, with the more liberal
models and assumptions common in evolutionary computation and
artificial life. There are, of course, many technical details of
genetics and natural populations that are often overlooked in
computational abstractions: e.g. mutation rates, genome structure,
and selection coefficients. But rather than focus on these details,
we will focus more on the impact that such details have had on how
the research questions and models of the two disciplines differ. For
example: Population genetics models sometimes disregard new
mutations and account only for changes in frequency of existing
alleles; Many models that do account for new mutations focus on
deleterious mutations, and/or neutral mutations, and less emphasis
is placed on beneficial mutations; Population genetics models also
tend to address a small number of loci and often ignore epistasis or
use simplified assumptions about epistasis. These types of models do
not merely reflect the different styles of different disciplines but
are motivated by biological observations and theoretical concerns
such as the maintenance of variation under blending inheritance, and
the inability of natural populations to select for sets of alleles.
By better understanding the empirical and philosophical grounding of
models used in evolutionary biology we can better understand how to
facilitate communication and cross-fertilization between, on the one
hand, artificial life and evolutionary computation, and on the
other, the evolutionary biology that inspired it.
Part I: A population genetics primer
-
Evolutionary biology historical perspectives:
-
Natural selection, the Modern Synthesis, Mass
action, Neutral theory
-
Basic questions focusing on dynamics of alleles:
-
Probability of fixation of an allele; time to
fixation of an allele
Part II: Epistasis and fitness landscapes
-
Definitions of epistasis and landscapes:
-
Synergistic, statistical, and unidimensional
epistasis; allele-frequency vs. genotype spaces
-
Contrasting assumptions in biology and computer
science
-
Magnitude vs. sign epistasis; assumption of
single-peaked landscapes
Part III: Sexual recombination
-
Simple models for the benefit of sex in population
genetics:
-
Fisher/Muller; Deterministic Mutation Hypothesis
-
Selection on combinations of alleles vs.
individual alleles
-
Shifting Balance Theory; The role of sex in
disrupting vs. creating combinations of alleles
