Theoretical and Experimental DNA Computation (Natural ...

134 5 Physical Implementations
We also carried out extensive control experiments to ensure the specificity
of the PCR detection step (i.e., to ensure that strands were not removed
without this being done explicitly). Also, control experiments indicated the
inefficiency of the Sau3A restriction enzyme that was originally used, though
an isoschizomer, MboI, worked well.
Ensure that the initial library is constructed cleanly before
proceeding
A fundamental prerequisite for correct algorithmic implementation is that
the initial library of strands be constructed as expected. This is especially
important for algorithms within filtering models, since we must be absolutely
sure that every possible solution to the given problem is represented as a
strand. While describing their attempt to recreate Adleman’s experiment,
Kaplan et al. [86] acknowledge the difficulty of obtaining clean generation of
the initial library.
There are several potential problems inherent to the construction of an
initial library by the annealing and ligation of many small strands. Incomplete
or irregular ligation can result in shorter than expected strands. We checked
for this, and observed that the majority of the product was of the expected
length. In addition to checking the length of the product, we rigorously ensured
that there is sufficient variability within the initial library by cloning a sample
into E.coli and sequencing their DNA.
Correct strand/primer design is vital
In [3] Adleman originally suggested using random sequences to represent vertices
within the given graph. He explained this choice by stating that it was
unlikely that sequences chosen to represent different vertices would share long
common subsequences, and that undesirable features such as hairpin loops
would be unlikely to occur. The selection of random sequences was also supported
by Lipton in [98].
Since the publication of [3] and [98], the use of random sequences has been
called into question [23, 51, 52, 108]. It is clear that for any nontrivial problem,
careful thought must go into the design of sequences to represent potential
solutions if we are to avoid the problems described above.
One major problem we encountered was due to the sequences chosen to
represent target sequences. We made a completely arbitrary decision to differentiate
S1 by the sequence AAAAAA, S2 by CCCCCC,andS3 by GGGGGG.
In retrospect, it is clear that this was a bad choice for two main reasons. The
first concerns the S2 and S3 primers. It is clear that, in solution, these primers
are complementary, and are just as likely to anneal to one another as they
are to the target sequences. Obviously, this will greatly reduce the efficiency
of the removal operation. The second problem concerns the melting temperatures
of the primers. Because the melting temperatures of the S1 primers was