CONTACT: Stanford University News Service (650) 723-2558
Progress in chemical-based computing reported
STANFORD -- John Ross' efforts just might lead to better computing through chemistry.
The Stanford chemistry professor has shown not only that certain biochemical reactions can form the basis for a device that actually uses chemical reactions to process information, but also that applying analytic approaches that were developed to study digital circuitry can provide valuable new information about complex biochemical reactions.
"It's just as interesting to take this in one direction as it is to take it in the other," Ross said in a speech at the American Association for the Advancement of Science meeting in Atlanta. In his presentation on Tuesday, Feb. 21, he summarized his recent progress in the study of chemical computation as part of a session that he organized on the subject.
Four years ago, Ross became interested in the question of how to do computing with chemistry, an area in which there has been considerable activity. A number of scientists have concentrated on using proteins or nucleic acids as the basis for molecular computing devices. Ross, however, asked a different question: Can chemical reactions themselves be used to process information?
"People get their energy by converting one-third of their body weight, about 25 kilograms, from adenosine diphosphate to adenosine triphosphate. About one-third of this chemical energy is used in the head. That's why your head gives off more heat than other parts of your body. So thinking may be at least partially a kinetic chemical process," Ross said.
Using various combinations of enzymatic reactions that take place within the cell, Ross, in collaboration with postdoctoral students Allen Hjelmfelt and Adam Arkin, has identified biochemical reactions that duplicate the basic logical functions from which virtually any computer can be constructed. In the August 1994 issue of the Biophysical Journal, he and Arkin identified several of the reactions involved in glycolysis -- the basic metabolic process by which sugars are broken down for energy -- that can act as logic functions.
One such function is the logical AND. Take the logical statement "Dick AND Jane," where Dick and Jane can either be present or absent. The statement has one value, call it true, if both Dick and Jane are present, but if either or both are absent, then it has a second value, false. The AND function can be duplicated with a biochemical system involving three compounds (A, B and C) and four enzymes (E1, E2, E3 and E4). The first enzyme converts compound A into B. The second converts A into C. The third converts B into A and the last converts C into A. In addition, there are two chemical inhibitors, one that turns off E1 and the other that shuts down E2. The concentrations of the two inhibitors serve as the inputs while the concentration of A is the output of this device. The concentration of A remains low unless both inhibitors are present, in which case the concentration jumps to a much higher level.
(Ross and his students also have shown that it is possible to simulate neural networks chemically as well as logically. They did so using chemical reactions that behave analogously to nerve cells. Such systems have a threshold. If the input signal is below this level, the signal dies away, but if it is above this level, then it is amplified.)
According to the researchers, there are two basic ways to connect such chemical elements into a computer.
One approach is to cascade the chemical reactions much as they are in the cell. The output of one set of reactions acts as the input to a second set whose output, in turn, acts as the input to a third set of reactions, and so forth. "These systems are of biochemical interest, but constructing a device based on these principles is a formidable challenge," Ross acknowledged.
The second, simpler way is to choose one or two chemical reactions, put them in separate chambers and connect them using a transport or diffusion process. In a paper that has been accepted by the European journal Physica D, Ross and Hjelmfelt propose using such an approach to build a simple pattern recognition system using one of the best studied but extremely slow bistable chemical reactions: the iodate-arsenous reaction that fluctuates between a high iodine state, which is blue, and a low iodine state, which is colorless.
"Chemical reactions tend to be pretty slow, compared to computers. Yet you can recognize me, and a thousand other people, in a tenth of a second. A tenth of a second is an eternity for a Cray supercomputer, but it can't do what you can do. So pattern recognition is one of the most likely applications for chemical computing," Ross said.
The Stanford scientists have collaborated with Jean-Pierre Laplante and Maria Payer of the Military College of Canada in Ontario to design and construct such a pattern recognition system. In an upcoming issue of the Journal of Physical Chemistry, they will report making an eight node system that can be interconnected with up to 32 sets of tubes and pumps.
The system has the capability of storing up to three different patterns, which are "hard-wired" into the system by the way in which the chambers are connected. An initial pattern is put into the system by filling each of the chambers with either high or low concentrations of iodine. The pumps are turned on. If the initial pattern is "recognizable" -- close enough to one of the stored patterns that it can be distinguished -- over a period of about 60 minutes the system stabilizes on the stored pattern. If the initial pattern is unrecognizable, the iodine concentration in all the chambers becomes homogeneous.
"Although we have examined the performance of [an eight node] network under a restricted range of conditions, both the powers and limitations of the experimental system agree with numerical simulations, and thus lend support to the validity of the simulations of larger networks," the authors concluded.
Most recently, Ross has begun to explore how to use this computational perspective to help unravel the complex biochemical reactions that occur within the cell.
"If we can build computers using chemical reactions, then we should be able to apply the tools that have been developed to analyze digital circuits to give us more information about complex chemical reactions," Ross said.
In an article published in the Journal of Physical Chemistry, he and Arkin show that it is possible to determine reaction mechanisms of model cascade reactions, similar to those involved in glycolysis.
"For complex reactions, chemists usually guess at the reaction mechanism and then test to see if they are right. But by taking a leaf out of the book of electronic circuit theory, we have shown that it is possible to deduce essential elements of this mechanism," Ross said.
If an electrical engineer is presented with a black box and asked to discover the circuitry hidden inside, he begins by sending electrical signals of different voltage, amperage and duration into the black box and then measuring the electrical signals that come out.
Ross' approach, which he calls correlation metric construction, works in a similar fashion. He systematically varies the chemical concentrations of each of the chemical compounds involved in a chemical reaction network. At each step, he measures the resulting concentrations of the other chemicals in the system. Using a series of mathematical techniques, he can then, in large part, reveal the underlying chemical reactions and the strength of the interactions between the different compounds.
This is an archived release.
This release is not available in any other form.
Images mentioned in this release are not available online.
© Stanford University. All Rights Reserved. Stanford, CA 94305. (650) 723-2300.