Test-tube stickers for DNA-based computers.

A vial holding a teaspoonful of DNA molecules looks tiny next to the most powerful of today's supercomputers. To some computer scientists, however, the 1020 or so strands of DNA contained in the vial represent the possibility of processing information at rates 1,000 times faster than those of any electronic computer presently available-if researchers can find a way of tapping the information-handling capability of individual DNA molecules. In 1994, computer scientist Leonard M. Adleman of the University of Southern California in Los Angeles demonstrated for the first time that standard test-tube manipulations of DNA can be used to solve a genuine, though simple, mathematical problem (SN: 11/12/94, p. 308). His method involved joining small DNA segments together to create longer molecules that represented possible solutions of his problem, then filtering out the one type of strand giving the correct answer.

Now, Adleman and his collaborators have developed, in theory, an alternative method of DNA-based computation. Unlike previously proposed methods, their so-called sticker model offers a way of storing and manipulating data, encoded as strings of 0s and 1s, using DNA molecules of fixed length and identity. "It's a new model of computation-a new way of representing information using DNA," says graduate student Sam T. Roweis of the California Institute of Technology in Pasadena. Roweis described the sticker model of computation at a meeting on DNA-based computers, held last week at Princeton University. A DNA molecule consists of two intertwined chains, each one a sequence of four simpler molecules, or bases, known as adenine, thymine, guanine, and cytosine. The four bases, designated by the letters A, T, G, and C, constitute complementary pairs: A sticks to T, and G sticks to C. These pairings hold two chains of bases together, creating the characteristic double helix of DNA. As the basis of their proposed molecular computer, Roweis and his coworkers use two different types of single-stranded DNA. Memory strands consist of long sequences of bases subdivided into segments of a standard length that represent individual memory units. Sticker strands are shorter-the same length as the memory units-and each sticker consists of a sequence of bases complementary to that of one of the memory regions.

When a sticker is "glued" to its complementary memory segment, the bit corresponding to that segment is on, representing 1. The absence of a sticker means that the corresponding bit is off, representing 0. A memory strand together with its stickers therefore represents a unique string of 1s and 0s. This is known as a bit string.

To solve mathematical problems, the researchers defined four different ways of manipulating these DNA bit strings, based on biochemical processes feasible in the laboratory. One fundamental operation combines two sets of bit strings. Another separates a combined set into its component sets. A third operation turns on a particular bit in a string, and a fourth turns off a certain bit. Appropriately combined, these operations can solve a variety of mathematical problems, Roweis says. However, he admits, "we've not actually got any of this to work in the lab yet."