BIOLOGICAL INSPIRATION FOR COMPUTING 289
Researchers have begun to construct cellular logic gates in which signals are represented by protein
concentrations rather than electrical voltages, with the intent of developing primitives for digital com-
puting on a biological substrate and control of biological metabolic and genetic networks. In other
words, the logic gate is an abstraction of an underlying technology (based on silicon or on cellular
biology): once the abstraction is available, the designer can more or less forget about the underlying
technology.
A biological logic gate uses intracellular chemical mechanisms, such as the genetic regulatory
network, metabolic networks, or signaling systems to organize and control biological processes, just as
electronic mechanisms are used to control electronic processes.
Any logic gate is fundamentally nonlinear, in the sense that it must be able to produce two levels of
output (zero and one), depending on the input(s), in a manner that is highly insensitive to noise (hence,
subsequent computations based on the output of that gate are not sensitive to noise at the input). That
is, variations in the input levels that are smaller than the difference between 1 and 0 must not be
significant to the output of the gate.
Once a logic gate is created, all of the digital logic design principles and tools developed for use in
the electronic domain are in principle applicable to the construction of systems involving cellular logic.
A basic construct in digital logic is the inverting gate. Knight et al.^122 describe a cellular inverter
consisting of an “output” protein Z and an “input” protein A that serves as a repressor for Z. Thus,
when A is present, the cellular inverter does not produce Z, and when A is not present, the inverter does
produce Z. One implementation of this inverter is a genetic unit with a binding site for A (an operator),
a site on the DNA at which RNA polymerase binds to start transcription of Z (a promoter), and a
structural gene that codes for the production of Z.
Protein Z is produced when RNA polymerase binds to the promoter site. However, if A binds to the
operator site, it prevents (represses) the binding of RNA polymerase to the promoter site. Thus, if
proteins have a finite lifetime, the concentration of Z varies inversely with the concentration of A. To
turn this behavior into digital form, it is necessary for the cellular inverter to provide low gain for
concentrations of A that are very high and very low, and high gain for intermediate concentrations of A.
Overall gain can be increased by providing multiple copies of the structural gene to be controlled by
a single operator binding site. Where high and low concentrations call for low gain, a combination of
multiple steps or associations into a single pathway (e.g., the mitogen-activated protein [MAP]-kinase
pathway, which consists of many switches that turn on successively) can be used to generate a much
sharper nonlinear response for the system as a whole than can be obtained from a single step.
Once this inverter is available, any logic gate can be constructed from combinations of inverters.^123
For example, a NAND gate can be constructed from two inverters that have different input repressors
(e.g., A1 and A2) but the same output protein Z, which will be produced unless both A1 and A2 are
present. On the other hand, cellular logic and electronic logic differ in that cellular logic circuits are
more inherently asynchronous because signal propagation in cellular logic circuits is based on diffusion
of proteins, which makes both synchronization and high speed very hard to achieve. In addition,
because these diffusion processes are, by definition, not channeled in the same way that electrical
signals are confined to wires, a different protein must be used for each unique signal. Therefore, the
number of proteins required to implement a circuit is proportional to the complexity of the circuit.
Using different proteins means that their physical and chemical properties are different, thus complicat-
ing the design and requiring that explicit steps be taken to ensure that the signal ranges for coupled
gates are appropriately matched.
(^122) T.F. Knight and G.J. Sussman, “Cellular Gate Technology,” Unconventional Models of Computation, C. Calude, J. Casti, and
M.J. Dinneen, eds., Springer, Auckland, New Zealand, 1998.
(^123) In general, the availability of an inverter is not sufficient to compute all Boolean functions—an AND or an OR function is
also needed. In this particular case, however, the implementing technology permits inverters to be placed side by side to form
NOT-AND (NAND) gates.