Purdue NewsPurdue News
November 1995
Venkat Venkatasubramanian, associate professor of chemical engineering at Purdue University, has applied Darwin's principles of evolution and natural selection to the computer-aided design of complex molecules that can be used to create new materials.
The program could be a boon to chemical, plastic and pharmaceutical companies that spend billions of dollars and thousands of man-hours trying to develop new drugs, new chemicals, or new petroleum products, he says.
"The physical and chemical properties of these materials are determined by the structure, composition and processing of their basic units -- molecules," he says. "To come up with one specific new molecule from scratch, you have to look at trillions of possible combinations. This program narrows those trillions down to a few possible candidates that have the particular physical properties you are looking for, such as a certain viscosity in a motor oil additive, or a certain hardness in a new type of plastic.
"By and large, new materials currently are developed by trial and error by experienced chemists and chemical engineers drawing on many years of experience trying various combinations. It's not very efficient, and it's very expensive. This computer program can dramatically cut down on the time it takes to design completely new products."
Venkatasubramanian presented the latest results of his research in October at a conference of the International Federation of Operational Research Societies in St. Louis.
Venkatasubramanian says he first got the idea for the computer program eight years ago by looking at how nature designs complex molecules.
"We are all products of nature's design," he says. "At the core, we are all composed of complex biological molecules, such as DNA. If you think about the principles of evolution and natural selection, they are essentially random processes, with the concept of 'survival of the fittest' guiding evolution. It made sense to me to try to mimic nature's success."
Venkatasubramanian's technique relies on a mathematical process called a genetic algorithm, which has been around for about 25 years, he says. Although genetic algorithms have been used in computer programs for scheduling, investment strategies and determining the 3-D structure of proteins, Venkatasubramanian is the first to apply them to computer-aided molecular design.
Here's how the process works: The first step is to determine the macroscopic, or bulk, properties you would like the new material to have, such as a certain level of viscosity and heat tolerance of a motor oil additive. This is the target molecule.
The computer program starts with a "soup" of 100 simple, randomly created molecules swimming around, conditions similar to those on the early Earth. The molecules then are allowed to "bump into each other," and in doing so react together chemically until 100 different molecules are formed. Some molecules break apart and recombine with parts of other molecules, resulting in large structural changes called crossovers. Smaller structural changes are called mutations.
The new molecules that are created form the first generation, or the offspring of the primordial parents. By looking at what each new molecule is made of and how it is put together, Venkatasubramanian can predict the properties these new structures have.
"I can determine how far away or how close each of these structures is to my target molecule," he says. "The program expresses this distance with a number on a 'fitness' scale from 0 to 100. 100 means I'm right on target, 0 means I'm way off. Typically I'm somewhere in between."
The program calculates the fitness of every offspring and eliminates those that are lowest on the scale, keeping only those that scored higher than 90, for example. The fittest molecules go on to interact with each other to produce the next generation, and so on.
"From generation to generation, I filter out the candidates by applying the 'survival of the fittest' idea," Venkatasubramanian says. "Hopefully the fitness will increase from generation to generation. Eventually, maybe by the 100th, 200th or 1,000th generation, I will have one or several molecules that are 100 percent fit.
"I say 'hopefully' because mathematically you cannot prove that each generation will necessarily improve in fitness. This is known as the convergence problem. You may start with very bright parents, but the kids are all stupid. It happens."
In practice, however, Venkatasubramanian and colleague James Caruthers, Purdue professor of chemical engineering, have found that in almost all the dozen or so cases they looked at, the program succeeded in locating the target.
In their most recent project, the researchers ran the program through 500 to 1,000 generations for each molecule, repeating each run 50 times in order to get a statistical average. In the 5 to 10 minutes it took for each run, they were able to find the target at least once for most of the molecules. Some they found very quickly, in only 40 generations. Some took 500 or more generations. In one case, they never found the target.
"Even when you don't find a 100 percent perfect fit, you don't go away empty handed," Venkatasubramanian says. "You may have a bunch of candidates that are 90 percent fit, and in some applications, 90 percent is good enough. It's like throwing darts. You may not hit the bullseye, but you still get points for hitting the circle around it."
Although he has not yet created any new materials in the lab, Venkatasubramanian says he has talked with colleagues in industry who may be interested in doing so in the next year.
Venkatasubramanian says the computer program does not require much computer expertise or special equipment to operate. He expects the program will be ready for industrial use by the end of this year. He currently is developing an interface that will make the program easier to use.
"The program is interactive, so I'm developing an interface that will allow you to watch the molecules evolve on the screen and interrupt the evolution if you see something interesting," he says. "At any point you can click the mouse on any molecule, manipulate it by replacing a chlorine by a bromine, for example, and let evolution continue from there. It's kind of like playing God."
Venkatasubramanian's research has been funded by the Purdue Research Foundation and industrial sponsors.
Source: Venkat Venkatasubramanian, (765) 494-0734; home, (765) 497-2473; Internet, venkat@ecn.purdue.edu
Writer: Amanda Siegfried, (765) 494-4709; home, (765) 497-1245; Internet, amanda_siegfried@purdue.edu
NOTE TO JOURNALISTS: Color photo of Venkat Venkatasubramanian in the lab is available from Purdue News Service, (765) 494-2096.