Biomedical Engineering Reference
In-Depth Information
Figure 2. GA gene structure
Figure 3. Example of GA genes for the encoding of cellular genes
The encoding of the individual genes follows
a structure that is similar to the one described
in Figure 2, where the number of promoters of
each gene may, vary but the white and indivisible
section “Activation Percentage - Constituent - Se-
quence” (PCS) must always be present. The PCS
sections determine the genes of the individual,
and the promoter sections are associated to the
PCS sections, as shown in Figure 3.
The search of a set of structures similar to
those shown in Figure 3 required the adaptation
of the crossover and mutation GA operations to
this specific problem.
Since the length of the individuals is variable,
the crossover had to be performed according to
these lengths. When an individual is selected, a
random percentage is generated to determine the
crossover point of that individual; for instance, in
Figure 4, the point that was selected as crossover
point corresponds to 60% or the length. After
selecting the section in that position, a cross-
over point is chosen for the section selected in
the other parent. Once this has been done, the
crossover point selection process is repeated in
the second selected parent in the same position
as in the previous individual; on the figure this
point corresponds to 25%. From this stage on,
the descendants are composed in the traditional
way, since they are two strings of bits. We could
execute a normal bit strings crossover, but the
previously mentioned steps guarantee that the
descendants are valid solutions for the DNA
strands transformation.
It should be highlighted that, by using this type
of variable length crossover, the number of gene
promoters, together with the change in sections
types, may induce the genes behaviour to vary
drastically when the new individuals are evalu-
ated. Another effect of this type of crossover is
the absence of promoters for a given gene: if this
occurs, the gene is either never expressed, or, if
it is constituent, never inhibited.
With regards to mutation, it should be men-
tioned that the types of the promoter or PCS
sections are identified according to the value of
the first string bit. Bearing that in mind, together
with the variable length of individuals, the muta-
tion operation had to be adapted so that it could
modify not only the number of these sections, but
also the value of a given section.
The probability of executing the mutation is
usually low, but this time it even had to be divided
into the three possible mutation operations that
the system contemplates. Various tests proved that
the most suitable values for the distribution of the
different mutation operations, after the selection
of a position for mutation, were the following:
for 20% of the opportunities, a section (either a
Search WWH ::
Custom Search