Biology Reference
In-Depth Information
collection of
genes
(Fig. 5). The mutation operator is one of the evo-
lutionary methods that can generate offspring solutions for the next
generation. Mutation is the only evolutionary operator in use in EP,
is the primary operator used in ES, and is a secondary operator in GA
applications. The implementation of mutations depends on the repre-
sentation. Mutation in binary representation presents problems by
selecting a bit and randomly changing it to 0 or 1 (independently of
the original value of that bit) or by inverting the value of bits selected
for mutation. Mutation in real-valued representations requires a ran-
dom change in the value of the real number encoded at a selected
position in the genotype. The crossover operator is the second evolu-
tionary operator used to generate offspring from selected parents.
Crossover is the primary operator in EP, a secondary operator in GA,
and not used at all in EP. Crossover combines genetic information
from two parents to generate one or two children that have features
from both parents. Crossover can be implemented in a number of
ways (Fig. 6(B)). The selection and the next-generation process
include a “survival of the fittest” mechanism. The computational
implementation of such a mechanism requires a way to calculate the
fitness of each member of the population (fitness function) and an
algorithm for selecting members of the population that will pass their
genes to the next generation. Fitness values should reflect the relative
quality of the problem being optimized to direct the evolutionary
search toward a more promising solution. Thus, fitness functions for
calculating such values must be written specifically for each problem.
As a result, the calculation of fitness values often consumes the major-
ity of the time dedicated to solving problems by EA. However, there
is no requirement for fitness values to be determined computationally.
GA implementation can use experimentally derived values determined
from biological assays that can serve as fitness values for population
members.
247
There are many types of evolutionary algorithms used
for the selection process, but their discussion is beyond the scope of
this chapter.
As previously mentioned, Brusic and colleagues (1998(b),
121-130) developed a hybrid method for the prediction of HLA-
DR4 (B1*0401)-restricted peptides. In this case, a GA method was
