The DNA molecules in our chromosomes are vast data banks full of vital information about the human body. US scientists have now shown that DNA can not only hold but also process information. They have used it as a chemical computer to solve one of the hardest of computational problems.

In essence, a strand of DNA encodes the information needed to put together protein molecules. This information is encoded in a string of four different molecular units. Short strands of synthetic DNA can now be made to order.

This data-bearing capability of DNA is the basis of a new type of computer. Tailor-made lengths of DNA that encode all possible solutions to a particular problem can allow researchers to weed out the 'wrong answers' in a massively parallel process rather than looking at each of them one by one.

Just about the meanest problems you can set a computer belong to the class called 'NP-complete'. The number of possible answers to these conundrums, and so the time required to find the correct solution, increases exponentially as the problem is scaled up in size. A famous example is the 'travelling salesman' puzzle, which involves finding the shortest route connecting all of a certain number of cities.

Problems like this are common in science and technology. Solving the travelling-salesman problem is a little like finding the most stable folded shape of a protein's chain-like molecular structure -- in which the number of 'cities' can run to hundreds or even thousands. Now, Lloyd Smith and colleagues from the University of Wisconsin, USA, have applied DNA computing to a related NP-complete problem, (called the 'satisfiability problem' or 'SAT')^{1}.

Sometimes the only available approach to solving NP-complete problems is to laboriously try out every possible route to see which of them is shorter. The beauty of DNA computing is that it does this sifting of possible solutions in parallel -- investigating many options at once.

The solution to a SAT problem has to satisfy several criteria simultaneously. Either this or that orthe other has to be true, at the same time as yet another orstill another. So every little part of it has to be checked in turn.

The group worked on a relatively small-scale SAT problem, with 16 possible solutions, all of which they encoded as a strand of DNA. To keep track of every sequence, Smith and colleagues fixed one end of each strand to a gold surface. They then began to eliminate the wrong solutions by using DNA's famous ability to form a zip-like, two-stranded double helix in which one strip is 'complementary' to the other.

They exposed the tethered lengths to free-floating strands complementary to those that satisfy one of the criteria met by the correct solution. These floating strands stuck to all the tethered ones that met this criterion, leaving single all those that contradict it. Smith's group then added an enzyme that destroys all single-stranded DNA. The surviving single strands were then regenerated simply by gently heating them to strip off their complementary partners.

The researchers repeated this process, successively exposing the immobilized DNA to complementary sequences satisfying each of the problem's criteria, and then destroying all single strands. Those strands that did not meet each part of the correct solution were systematically eliminated en masse (rather than one by one, as a conventional computer would do). The remainder -- encoding the right answer -- were then multiplied and their sequences decoded.

Using this form of DNA computing to solve larger problems remains difficult, partly because pairing of strands does occasionally incur errors and partly because it is costly to make longer strands that encode more information. But it illustrates that, in principle at least, molecules can be smarter than any number of electronic switches.