Skip to main content

Big Red 200 powers quantitative approach to drug discovery

Michael T. Robo, staff scientist at the Indiana Biosciences Research Institute, works to prioritize potential molecules for drug development using complex models on IU’s Big Red 200 supercomputer

News and events Feb 28, 2024

Drug discovery is often compared to finding a needle in a haystack. The needle, in this case, refers to a molecule of 50,000 atoms moving and vibrating. Among billions of potential molecule compositions and three-dimensional forms, Michael T. Robo, a computational chemist on staff at the Indiana Biosciences Research Institute (IBRI), uses GPUs on Research Technologies’ high performance computing system Big Red 200 to zoom in on promising molecules for drug discovery.

Molecules already identified as useful drugs in medical treatment make up the first-order logic, or data archetype of this research. If a molecule is successfully used in breast cancer treatment, for example, Robo would start there and computationally tinker with its form. These initial models are called molecular docking and shape-based screening, which look for significant points of interaction between a molecule and a biological model like a protein. The models, usually run on CPUs, run millions of combinations of drug molecules and targets based on the form and static composition of the molecular structure.

Drug Molecules

Because these molecules are in constant motion, the next layer of Robo’s analysis considers the form and speed of the molecule through time, called Molecular Dynamics. “In classical physics, you start with all of these assumptions that don’t account for things like friction or gravity,” said Robo of the molecular docking model process. “The reason why you have these assumptions is that it simplifies the amount of work you need to do to address your hypothesis within a limited framework. But when you want to expand the scope of your framework, you need additional computing power.”

The molecular dynamics model has a high computational cost of figuring out the direction of the molecule’s constant movement. “You can calculate the forces of each atom in the system, and usually there’s about 50,000 atoms in the system,” he said. “From those, you can figure out the momentum and their position after a short time step. But because you need to do that over and over again and you need to take 2.5 million time steps, you really need GPUs.”

Michael Robo Michael Robo, a computational chemist on staff at the Indiana Biosciences Research Institute (IBRI)

“With Big Red 200 and this multilayered model, we can look at millions and billions of molecules a day,” said Robo. “My job is to use this information to guide the IBRI’s choices in purchasing compounds or move beyond that and synthesize a compound in the lab. One compound takes about a week to synthesize, so If you can take 1000 ideas and build a model that prioritizes 10, it potentially saves weeks of work.” Robo’s portfolio includes work on neurofibromatosis 2 and Alzheimer’s disease research with the IU School of Medicine.

Robo, who completed postdoctoral research in computational chemistry under Jonah Vilseck, said he’s benefitted from the high level of expertise provided by the engineers at Research Technologies. “My postdoc research considered the role of temperature and thermodynamics in drug discovery simulations using a technique called replica exchange. It required a lot of processes running at once as several GPUs were talking to each other,” he explained. “As a computational chemist, I’m thinking about my experiment and not necessarily about optimizing the code or the order in which you compile programs—this is where the RT team has been very helpful,” he said.

Find out more about Big Red 200 GPUs:

More stories