A Design for DNA Computation of the OneMax Problem
David Wood Junghuei Chen Eugene Antip ov Bertrand Lemieux Walter Cedeno ~
Computer Sci. Chem. & Bio chem. Chemical Eng. Plant & Soil Sci. Hewlett-Packard
Univ. of Del. Univ. of Del. Univ. of Del. Univ. of Del. Centerville Rd
Newark DE Newark DE Newark DE Newark DE Wilmington DE
Abstract details of the lab oratory op erations, and some prelimi-
nary lab oratory results can b e found elsewhere [5, 38].
Elements of evolutionary computation and
1.1 Conventional Evolution Computation
molecular biology are combined to design a
DNA evolutionary computation. The tradi-
Evolution Computation usually maintains a p opula-
tional test problem for evolutionary compu-
tion of candidate solutions, often represented as strings
tation, OneMax problem is addressed. The
of binary bits. In each generation, some less promising
key feature is the physical separation of DNA
candidates are likely to b e eliminated. Their replace-
strands consistent with OneMax \ tness."
ments are generated from relatively promising candi-
dates by cloning and/or mutation and/or recombina-
tion. Evolutionary Computation comes in many, many
1 Intro duction
di erentvarieties [3, 19, 20], but most are variations
on this outline:
Evolution is a concept of obtaining adaptation through
the interplay of selection and diversity. Analogies from
Begin with an initial p opulation, often chosen ran-
evolution have b een used in b oth conventional comput-
domly.
ing and molecular biology. In computing, the general
term is \evolutionary computation." The paradigm in
Rep eat the following steps until convergence.
molecular biology is known as \in vitro evolution."
1. Evaluate tness of candidates.
In this pap er we identify elements of evolutionary com-
2. Select candidates to breed, favoring the more
putation and molecular biology that we combine to
t over the less t.
address the traditional OneMax problem. Our design
3. Delete some of the less t.
can b e mo di ed to address some other problems as
4. Breed replacements, inducing variation.
well [5], as is brie y discussed in Section 5. Wecho ose
evolutionary computations that manipulate bitstrings
using op erations of p ointwise mutation and crossover.
2 DNA Suits Evolutionary
These op erations can b e p erformed by mo di cations
Computation
of techniques from molecular biology.
Section 4.1 forms the heart of this pap er. The cru- Several means of DNA computation have b een ad-
cial lab oratory op eration is physically separating DNA dressed. The rst was, of course, by Adleman [1, 2].
strands by their \ tness" for solving the OneMax prob- Recentoverviews can b e found in [17] and [22]. See
lem. In this section, we identify two-dimensional de- also the DNA computing bibliography of Dassen and
naturing gradient gel electrophoresis as an appropriate Pierluigi [9].
lab oratory technique for this crucial separation.
From the b eginning of DNA based computing to the
We are careful to make the following p oint. In this pa- present there have b een calls [11,24,29, 6] to consider
per we only aspire to provide the design for some evo- carrying out evolutionary computations using genetic
lutionary computations using p opulation sizes larger materials in the lab oratory. So far, there have b een
than is practical with conventional computers. More three such exp eriment designs prop osed, including two
one must physically separate DNA strands according in a recent DIMACS Workshop [4,5]. The very rst
to their \ tness" for solving the problem at hand. design was presented ab out twoyears ago [10], but has
not yet b een carried out in the lab oratory.
2.2 DNA Evolutionary Computation
With our design, computing time using DNA is pro-
Compared to Sup ercomputers
p ortional to the numb er of generations. This mo-
tivates incorp orating b oth p ointwise mutation and
The following oversimpli ed estimates indicate DNA
crossover, attempting to minimize the numb er of gen-
computing techniques could in the future compare
erations required.
favorably to sup ercomputers in some cases. Favor-
For that matter, one mightwant to consider any
able cases include executing evolutionary computa-
additional evolutionary analogies that might reduce
tions having very simple tness evaluations and ex-
the numb er of required generations. These might in-
tremely large p opulations of candidate solutions.
clude transp osition, inversions, introns, etc. These
Consider a p opulation represented by a total of p bits.
have b een included in researchinevolutionary com-
Executing a evolutionary computation by any means
putation using conventional computers. Computing
will require at least g Op tness evaluations, where
paradigms inspired byevolution evolutionary compu-
g is the numb er of generations used. Now, assume
tations seem particularly suited to implementation us-
the tness evaluation of a candidate solution pro cesses
ing DNA. This is b ecause evolutionary computations
all the bits of the candidate solution. A state-of-the-
often use bitstrings, crossover, and p ointwise muta-
12
art tera op sup ercomputer p erforms ab out 10 op er-
tion, all of which are done with DNA in nature.
ations p er second. Thus, wehave a rough estimate for
sup ercomputer time complexity,
2.1 DNA Advantages for Evolutionary
Computation