14 September 1998
The New York Times
Four years ago, Dr. Leonard Adleman raised the exciting possibility of DNA-powered supercomputers when he wrote in the journal Science that he had solved a computational problem essentially by stirring up some DNA in a test tube, using building blocks of DNA as computing symbols.
The technique, in concept, involves the use of DNA -- strands of genetic code -- as a stand-in for computer software code.
Adleman, a professor at the University of Southern California, set up his problem by synthesizing DNA in a certain sequence and then letting DNA molecules react in a test tube so they ultimately produce a molecule whose sequence is the answer to the problem.
He used the chemical units of DNA rather than electronic 1s and 0s to solve a single, relatively simple problem: figuring out the shortest distance someone would have to travel to visit a number of different cities -- similar to a standard math problem referred to as the "traveling salesman tour problem."
But last week Dr. Warren Smith and Dr. Allan Schweitzer, two scientists at the NEC Research Institute in Princeton, N.J., received what appears to be the first patent covering a DNA computing process that at least theoretically can solve not just a single problem but a broad array of problems, just as a conventional desktop computer might.
The patent describes a way to concoct a vat of DNA molecules that have been manipulated by standard biotechnology techniques -- such as the cutting, splicing and growing of DNA strands -- to behave like miniature computers.
All these molecular-size computers would operate in parallel, with each one exploring different possible computational paths to solve a problem, potentially speeding up computing time greatly.
The patent details a way to use genetic material to construct a large number of what are called Turing machines, all running in parallel, with each exploring different possible computational paths. (A Turing machine is a theoretical construct, first proposed in 1937 by the British computer scientist Alan Turing, that provides the schematic outline upon which digital computers are based.)
Despite the new patent, Smith does not sound optimistic these days about the patented technique.
"Some of the DNA molecules do the right step," he said in an interview, describing the process. "But then some other fraction just dissolve and get flushed down the drain."
This and other problems lead to an exponential buildup in errors during the calculations. In addition, the operations of this DNA computing so far are slow-going, taking hours.
"This does appear to be the first patent to issue in this area," said David Waltz, an NEC vice president. "But will this be the basis of a new industry, and will everyone have to take out a license on it in the future? That's not so clear."
Smith and Schweitzer received patent number 5,804,373.
[Balance of Patents column omitted]
Patents are available by number for $3 from the Patent and Trademark Office, Washington, D.C. 20231.
Copyright The New York Times