Information Technology Reference
In-Depth Information
checking in the first amino acid of the protein chain as well as starting it from
the center of the chain. The second repair strategy, with absolute moves, always
begins from the first amino acid. These three possibilities were used with the
same probability when the candidate solution used by DE generates an illegal
protein conformation.
Figure 2 shows a compari-
son between the two alterna-
tives, using DE with and with-
out the repair processes. The
protein series used was one of
the benchmark sequences shown
in Table 1 (sequence S2 ). The
quality evolutions in the Fig-
ure are an average of 5 differ-
ent evolutions with different ini-
tial populations. As the repair
processes allow that almost all
the individuals in the popula-
tion are feasible, DE combined
with them obtains the optimum configuration in fewer generations. DE with-
out the repair processes needs much more than the 100 generations shown in
the Figure to obtain the best configuration in all the runs. For the comparison
purpose, we also included in the Figure the evolution of the average quality of
the best individual of 5 different runs of a GA with typical parameters (one-
point crossover with probability of 0.5, mutations with a probability of 0.02,
tournament selection with a size of 4% of the population, and use of the relative
encoding).
generations
1
11
21
31
41
51
61
71
81
91
101
0
Best individual using repair processes
Average fitness using repair processes
Best individual without repair processes
Average fitness without repair processes
Best individual, GA
-1
-2
-3
-4
-5
-6
-7
-8
-9
-10
Fig. 2. Comparison of DE evolution with and
without the use of the repair processes
In Table 2 we included the comparison of our
hybrid DE solution with the results of Unger
and Moult with a GA [23], using the bench-
mark series for the square lattice. In the table,
E min represents the maximum number of HH
contacts. Previous works summarize the results
taking into account independent runs for each
proteinsequence,andineachcolumnitisspec-
ified the best energy value found in the different
runs of the corresponding search algorithm. If
the best known solution quality was found, it is
indicated in bold. The values in parentheses are
the minimum number of conformations scanned
before the lowest energy values were found in
one of the runs. This is the case of the results of
Unger and Moult [23] from 5 independent runs.
In our case, we included the average energy in
the different runs and the average number of
( S6)
(S5) (S8)
Fig. 3. Optimal configurations
for three of the sequences: S5,
S6 and S8
Search WWH ::




Custom Search