A collaboration of researchers from New York University and Duke announced this week the first successful computations performed by DNA molecules predesigned to self-assemble into structures that carry out computer logic calculations.

"This is the first time that anyone succeeded in getting a computation done by self assembly," said John Reif, a Duke professor of computer science who, together with Thomas LaBean, a Duke visiting assistant professor of computer science, and NYU chemists Nadrian Seeman and Chengde Mao described the accomplishment in the Sept 28 issue of the journal Nature.

"This was a very small-scale, modest experiment," Reif stressed in an interview. "But it really did work. Now we're trying very hard to do a large-scale experiment." His ultimate goal is trillions of separate computer operations at the same time by myriads of these self-assembling DNA structures, each known as a "tile."

Such a blinding performance by hordes of molecular assemblages individually only billionths of meters in size could beat today's best massively parallel supercomputers, Reif predicted. "The number of operations done in, say, 20 minutes might be many trillions beyond what silicon chip technology can do," he said.

The research has been supported by $2.7 million from the National Science Foundation and Defense Advanced Research Projects Agency under grants administered by Reif that have supported "biomolecular computing" work at 10 different universities. While that funding is now ending, the NSF recently announced another $2 million award to Reif and researchers at NYU, Duke and the University of South Carolina to continue explorations into self-assembling DNA computation.

DNA, or deoxyribonucleic acid, is located in the chromosomes packaged within each plant and animal cell. Each complex DNA molecule usually consists of two interactively spiraling strands that are held together by mutual attraction. They attract because each long strand is made up of combinations of the same four different chemical bases - abbreviated as A, T, G, and C - with any A on one strand naturally bonding to any T on the other, and likewise Gs to Cs. Also because of that attraction between its bases, DNA pieces have a natural tendency to self-assemble.

Molecular self-assembly is a special interest of Seeman's chemistry lab at NYU. Another collaborator, Erik Winfree of California Institute of Technology, in turn suggested that DNA molecules might be made to self-assemble into special tiles that could do computing. Winfree was drawing on the work of theoreticians in the 1960s who proposed that combinations of domino-like tiles could be assembled into computation machines by matching up patterns on the tiles' surfaces.

The molecular equivalent tiles are made up of tailored arrangements of DNA strands that cross over each other to form stable, vaguely rectangular shapes. These DNA tiles also have "sticky" ends made up of tailored combinations of A,T,G and C. Because A's on one tile will stick only to T's on another, and G's only to C's, only certain base-matched tiles will link together with other tiles. The "right" combination of DNA tiles could congregate into self-assembling complexes that perform computations in the same manner as domino tiles.

It was Reif who provided the mathematical rules of thumb, or "algorithms," on how to assemble the DNA tile combinations that performed the various basic computer logic calculations done in the experiment detailed in Nature. LaBean, a structural biologist, collaborated on tile design and construction in the laboratory, using some of the same methods employed in conventional DNA testing.

The Duke-NYU collaborators' Nature article described success in designing DNA strands that triple-cross-linked themselves into seven DNA tiles that solved simple Boolean logic computer operations of the kind known as XOR.

Computers calculate using only two integers, 0 and 1, by feeding in input data to produce output data - the results. XOR rules stipulate that if two input integers are the same, either 00 or 11, then the output will be 0. Conversely, if one input integer is a 0 and the other is 1, then the output will be 1. Here various DNA tiles took on the role of input and output integers.

In another study, Reif and LaBean showed how more-intricate XOR calculations could be used in cryptography, the science of creating (and breaking) secret codes. "You can get an encoded sequence that is absolutely unbreakable," Reif said.

Reif and other researchers are especially interested in using DNA for the kinds of computing problems that require large numbers of simultaneous calculations. Today's supercomputers tackle such challenges by assigning different pieces of the problem to different processing units that work in parallel. But Reif estimated that a quart of weak DNA solution could contain a million trillion different DNA strands, each the equivalent of a computer processor.

Contemporary supercomputers, assembled out of ultra-dense silicon-based integrated circuits, might achieve close to one trillion operations per second, while the DNA equivalent might perform one thousand trillion with much less energy consumption than current electronic machines, he noted.