Figure 5: An abstract model of a DNA complex can be created by first identifying
the atomic units of reaction, and then letting each serve as a node in a graph.
designing artificial biological systems
requires more flexible disciplines that
can cope with unexpected interactions
of components or errors caused by stochastic behaviors.
Life has solved these problems
through the power of evolution over billions of years. The introduction of evolutionary methods for designing artificial biological systems may therefore
be a reasonable strategy. Bio-inspired
evolutionary methods such as genetic
algorithms have been investigated for
decades and are now established as
useful engineering options for designing complex systems, including artificial genetic networks [ 3].
To this point, we have briefly explained the kinds of IT that can be
used for constructing artificial biological systems. Constructing molecular
systems made of DNA in the field of
DNA nanotechnology also requires
similar methods for modeling, simulation, and design.
A
CGCACAGT
A
CGACTT
b
b
a
GCGTGTCA
a
GCTGAA
CGCACAGT
B
Reaction
Units
A
Graph
B
A
“Imagining the
use of computers is
easy in the context
of construction
of automobiles or
other computers.
But imagining the
use of computers to
make robots made of
cells or DNA is much
more difficult.”
For example, the last step in designing DNA systems is to allocate a specific base sequence to each segment in
a DNA structure. This step is usually
called sequence design. This is a kind of
combinatorial optimization, because
the intended structure should be the
most stable of all the possible structures
formed by the designed DNA sequences.
Therefore, sequence design requires the
prediction of the free energy (i.e., stability) of each possible structure, and the
selection of the structure with the minimum free energy. Both problems have
been the subject of decades of investigation in the field of bioinformatics,
and powerful computer programs have
been developed for their analysis. For
example, Mfold is the best-known program for calculating the minimum free
energy of DNA or RNA [ 10]. In the case
of RNA, this requires a sophisticated dynamic programming method.
Even if determination of the struc-
ture with the minimum free energy is
possible, we should be able to design
sequences whose minimum free energy
structure is the intended one; this usual-
ly requires definition of the “goodness”
of sequences as a metric. In addition to
the free energy of the intended structure,
such a metric usually takes into account
a number of other parameters to satisfy
experimental conditions on sequences.
Sequences are then sought on the ba-
sis of an appropriate metric. This is a
typical optimization problem for which
bio-inspired evolutionary methods such
as simulated annealing, genetic algo-
rithms, and stochastic local search are
rather useful. Many programs for se-
quence design have been developed thus
far, including the program developed
by one of the authors [ 8], with no single
one manifesting as clearly better than
the others. Sequences designed by such
a program are usually validated by a free
energy program such as Mfold.