(Atlantic Books, £18.99)
Computers aren't what you think they are. Rather than plastic boxes filled with transistors, they could be made from ping-pong balls rolling down plastic tubes. Or, as Martyn Amos explains, they could be test tubes full of DNA.
Computation is about encoding and manipulating information, an abstract notion that makes no reference to any particular physical embodiment. A farmer who opens his gate to heifers but not to Friesians is doing computing—not just in a colloquial sense, but according to the formal definition supplied by the fathers of computational theory. Computers as we know them acquire their power not from any sophistication of their basic processes or components, but from the sheer speed and number of them.
It was Alan Turing who introduced this almost surreal view of computing in 1936, when he presented the notion of what is now known as the universal Turing machine. This is a device that stores and enacts an arbitrary set of symbolically encoded instructions. A universal Turing machine can carry out every feasible computing operation, but the hardware is immaterial. Turing imagined encoding the instructions using binary code—a string of 1s and 0s—on a tape, which could be magnetic recording tape or a paper strip with punched holes.
Or a code-carrying molecule. Given the similarity between Turing's tapes and the information-laden double helix of DNA discovered by Francis Crick and Jim Watson in 1953, it is surprising that no one thought of using DNA for computing until four decades later. By then there was plenty of talk about "molecular computers," driven mainly by the awareness that the miniaturisation of silicon-chip circuitry was approaching fundamental limits. This miniaturisation has driven the huge growth in computer power over the past half-century, but it cannot continue indefinitely because ever-smaller silicon circuits will eventually get leaky. One alternative is to make circuit components from single molecules.
But DNA computing, as first proposed in 1994 by Len Adleman of the University of Southern California, doesn't merely reproduce silicon circuits using molecules. Adleman's algorithm is not a linear sequence of logical steps that transforms input data into output data—a question into an answer. It entails making DNA molecules that encode all possible answers, then sifting them for the correct one. It works because DNA is so small. A test tube can contain an astronomical number of DNA molecules, and it is relatively easy to shuffle their chemical structures so as to encode many different "answers."
Adleman coupled his insight to practice, using DNA computing to solve a mathematical puzzle about finding the most efficient path connecting several points on a graph. On hearing about Adleman's paper, Princeton computer scientist Richard Lipton admitted that "suddenly I didn't know what a computer was any more."
Martyn Amos was one of many young computer science researchers captivated by this work, and in 1997 he became the first person to be granted a PhD in DNA computing. His story therefore has a personal flavour—for better and worse. While he generally makes a hard subject accessible, he adheres to the pop science template by including lots of character sketches of marginal relevance. When he reaches, say, the controversial chemist Kary Mullis, who suggested that he had been abducted by aliens, you sense the impending dollop of character profile with a sinking feeling.
Will DNA computing ever find its way into laptops? Amos skirts around this question. He admits that "traditional computers will still be important in our everyday lives for the foreseeable future," but argues that DNA computing is about rethinking the fundamental basis of computing. With its all-at-once ("massively parallel") operation, it subverts the one-thing-at-a-time ("serial") design of today's silicon chips. Quantum computing promises the same thing by different means: DNA computing puts all the possible solutions in the same flask, while quantum computing generally aims to put them on the same atom. (My bet is that quantum computing will find a viable form sooner than DNA, despite the technical challenges.)
But while this parallel approach could be useful for data searches, say, or factorising or encryption, it isn't suited to most computing problems. So Amos rightly suggests that the real niche for DNA computing may be biomedical, with DNA performing its usual role of encoding proteins. In 2004 a team from Israel demonstrated such a "DNA computer," which converted a biological input signal—the presence of molecules denoting disease—into an appropriate response, such as switching on the production of DNA-like drug molecules. This brings us full circle to DNA as a life-code—but this time it can be synthetic, its genetic repertoire invented from scratch. That could provide the basis of entirely new types of organism, in the discipline of synthetic biology. Call it computing if you will, but as Amos would be the first to admit, no bacterium is ever going to run Windows XP.
As a computer scientist, it is natural that Amos should focus on the evolutionary pathway on which the fundamentals of computing science intersect with the notion of DNA as information carrier. But there is another route, leading from chemistry. DNA replicates and makes proteins via "molecular recognition," a fitting-together of molecules like lock and key—or like complementary patterns of information. This stimulated Nobel laureate Jean-Marie Lehn to coin the idea of a science of programmable ("informed") matter in the late 1980s. Following this trail helps to strip away any suggestion that there is something unique about DNA as an informational molecule. Using DNA as a programmable molecule makes sense, because we can bring to bear the vast armoury of biotechnology for making, linking, cutting and pasting. But the idea is more fundamental than that. Once you see chemistry as an information science, DNA and all of biology becomes merely a particular example, much as the Apple Mac is just a Turing machine with great styling.
