[If you cannot see the flash video below, you can click here for a high quality mp4 video.]
Produced by Chris Bergendorff
Gaming the System
These days, video games are everywhere, from games that keep you fit to games that teach you to perform surgery. Video games have even invaded the realm of science, like in the recent hit computer game "SPORE," which lets players guide the course of evolution from single cell to galactic empire. But this isn’t just a one-way street. In turn, researchers at Michigan Technological University have found a way to use the computer graphic cards that power video games for biological research.
“You can use the graphic cards that were originally built for playing video games,” says Roshan D’Souza , a mechanical engineer at Michigan Tech, “and turn them around, and turn them into supercomputing devices to solve scientific problems.”
Also on ScienCentral
D’Souza originally began using these graphic cards to solve engineering problems. Often, after constructing the components of a device or machine, engineers discover that the pieces don’t work smoothly together, due to a problem with the original design. Much time and effort is wasted before an engineer can redesign and rebuild these parts. D’Souza found he could model the structure of components on a computer before building them, so as to get a sense of how they would fit together when finally assembled.
After using these powerful graphic cards, called graphics processing units, or “GPU’s”, for work in his own field, D’Souza realized this powerful technology might also benefit other areas of science. He asked his students to help him think of other applications for these GPU’s.
As D’Souza describes, "We came upon a paper in Computational Biology, that talked about bacterial colonies, and modeling bacterial colonies using agent based models."
Agent Based Modeling
Agent based modeling is a kind of computer modeling that seeks to replicate the behavior of large groups of independent ‘agents’ in a network or system. Some examples of systems that include lots of autonomous agents include cars in traffic, the path of infections in epidemics, even the behavior of consumers shopping for goods. The problem with agent based models is that they require a huge amount of processing power to simulate the behavior of all of the hundreds, thousands, or even millions of individual agents– power that normally isn’t possible on a typical home or office computer.
Enter GPU’s. These game processors are already powerful enough to model the behaviors of lots of individual characters in a computer game. If it works for agents in a game, D’Souza and his students reasoned that it should work for agents in real life systems as well.
So D’Souza and his students tried simulating a particularly complex biological system, in this case the behavior of a bacterial colony. After a few months of fine-tuning the programming, they created a simulation that seemed to be not only accurate, but also incredibly fast. When they compared their work with what other researchers around the world had accomplished, they found that the GPU’s were outstripping the competition.
As D’Souza puts it, “We found that we were nearly three orders of magnitude faster than any system that is out there.”
Faster, Stronger, Better
D’Souza and his team then demonstrated their work at the 2007 Agent Conference. There they showed that complex biological systems, like a tuberculosis infection, could be modeled on a simple personal computer, at a speed approaching that of multi-million dollar supercomputers. Such supercomputers can only be found in select research centers around the country, and access to them is very limited. By making use of these video game processors, researchers can run simulated experiments over and over again on their own office computer, in a fraction of the time it takes today.
"Something that takes maybe a month to compute on a traditional computing system, just divide that by a thousand," says D’Souza.
D’Souza says the simulations could also speed up the process of drug discovery. Pharmaceutical companies often have to waste effort and time on expensive clinical trials to test drugs that will ultimately not work for any number of reasons. But, with the use of GPU’s, the effects of different drugs can be simulated on a computer at less cost and greater speed. Instead of searching for the right drug among hundreds or thousands of the wrong ones, researchers can narrow down their search to just those drugs that work in the simulations.
"Once you narrow down your search space, then you can do clinical trials within that small search space. What that does is it greatly speeds up the development of new drug therapies,” says D’Souza.
Giving future scientists and doctors the same computational edge that gamers already enjoy.
Elsewhere on the Web:
Michigan Technological University – Manufacturing Computations Lab