Theoretical and Experimental DNA Computation (Natural ...

140 5 Physical Implementations
process was repeated three more times for the remaining three clauses, leaving
a surface containing only strands encoding a legal solution to the SAT
problem.
The remaining molecules were amplified using PCR and then hybridized
to an addressed array. The results of fluorescence imaging clearly showed four
spots of relatively high intensity, corresponding to the four regions occupied
by legal solutions to the problem (S3,S7,S8, andS9).
Although these results are encouraging as a first move toward more errorresistant
DNA computing, as the authors themselves acknowledge, there remain
serious concerns about the scalability of this approach (1,536 individual
oligos would be required for a 36-bit, as opposed to 4-bit, implementation).
5.9.3 Gel-Based Computing
A much larger (20 variable) instance of 3-SAT was successfully solved by Adleman’s
group in an experiment described in [33]. This is, to date, the largest
problem instance successfully solved by a DNA-based computer; indeed, as
the authors state, “this computational problem may yet be the largest yet
solved by nonelectronic means” [33].
The architecture underlying the experiment is related to the Sticker Model
(see Chap. 3) described by Roweis et al. [133]. The difference here is that only
separation steps are used – the application of stickers is not used. Separations
are achieved by using oligo probes immobilized in polyacrylamide gelfilled
glass modules, and strands are pulled through them by electrophoresis.
Strands are removed (i.e., retained in the module) by virtue of their hybridizing
to the immobilized probes, with other strands free to pass through the
module and be subject to further processing. Captured strands may be released
and transported (again via electrophoresis) to other modules for further
processing.
The potential benefits of such an approach are clear; the use of electrophoresis
minimizes the number of laboratory operations performed on
strands, which, in turn, increases the chance of success of an experiment.
Since strands are not deliberately damaged in any way, they, together with
the glass modules, are potentially reusable for multiple computations. Finally,
the whole process is potentially automatable, which may take us one step further
towards a fully integrated DNA-based computer that requires minimal
human intervention.
The problem solves was a 20-variable, 24-clause 3-SAT formula Φ, witha
unique satisfying truth assignment. These are