Harnessing Supercomputing Power for Drug Discovery
Summary
- A collaboration between the University of Miami’s Frost Institute for Data Science and Computing and a precision medicine company has proven that using powerful supercomputers can speed early-stage drug research.
- With the use of ALAFIA’s AIVAS Supercomputer, the Schürer Systems Drug Discovery Lab performed the simulations necessary for creating new HIV drugs up to 10 times faster than with standard methods.
- The approach cut the average time for simulations from days to hours, said Dr. Stephan Schürer.
A collaboration between the University of Miami’s Frost Institute for Data Science and Computing and a precision medicine company has proven that using powerful supercomputers can speed early-stage drug research.
With the use of ALAFIA’s AIVAS Supercomputer, the Schürer Systems Drug Discovery Lab, run by University of Miami Miller School of Medicine molecular and cellular pharmacologist Stephan Schürer, Ph.D., performed the simulations necessary for creating new HIV drugs up to 10 times faster than with standard methods. What’s more, the collaboration paves the way for future applications in cancer and other types of medical research.
Molecular Dynamics and Supercomputers
Like many collaborations, this one started by happenstance. Joey Schulz, a fourth-year Ph.D. student in biochemistry and molecular biology at the Miller School, was chatting one day with his parents’ neighbors. When he told them about his molecular simulation work, they offered to introduce him to their son’s best friend, Camilo Buscaron, cofounder of ALAFIA.
Schulz and Buscaron discovered a way their respective work could be mutually beneficial. As with many artificial intelligence (AI) programs, ALAFIA’s supercomputers rely on powerful processors called graphical processing units (GPUs).
Schulz’s research leverages molecular dynamics simulations. The computational approach is highly dependent on the power and parallelism of modern GPUs. Coincidentally, the performance benchmark widely used to rank the world’s fastest supercomputers is itself derived from a molecular dynamics code.
“The gold standard for benchmarking high-performance computing systems, including GPUs, is LINPACK or HPL. But many systems also use HPCG and molecular dynamics-based codes like HPL-AI and the AMBER or GROMACS benchmarks,” Schulz explained. “In my research, I simulate how a small molecule interacts with a protein target by modeling the physical forces between atoms over time. These simulations are extremely GPU-intensive because I need to compute how even femtosecond-scale atomic movements propagate through the entire molecular system, affecting binding, stability and conformational changes. The physics and math that make molecular dynamics so demanding are exactly what make it an ideal stress test for modern supercomputers.”
Schulz persuaded Buscaron to use Schulz’s real molecular dynamics data rather than the simulated data of the standard benchmarking tool. Schulz introduced Buscaron to Dr. Schürer, a Miller School professor of molecular and cellular pharmacology who directs the Frost Institute’s Digital Drug Discovery program. The collaboration was born.
Speeding Drug Discovery Research
The first computational task assigned to the ALAFIA supercomputer was to rank a panel of 25 candidate molecules to identify the most promising compounds for degrading HIV proteins. Schulz designed and implemented the custom computational pipeline used to evaluate each compound. The molecular data for this analysis was provided by a collaborator at the University of Alabama at Birmingham.
The idea worked.
Before, the team could only run four simulations at a time, for about 24 hours on average. With the ALAFIA supercomputer, the simulations take four hours max.
“It’s much more streamlined,” Schulz said.
Dr. Schürer added, “By adopting the AIVAS supercomputer, we cut run times from days to hours. The faster data processing, complex modeling and large-scale simulation accelerated breakthroughs in next-generation HIV treatments.”
The success of the experiment on HIV drug research offers promise for future collaborations on different types of medical research.
“The experiment simulated how a drug binds to a protein, something that underlies many therapeutics and infectious diseases. It could work similarly for targeted cancer therapeutics or other diseases. It’s all just different proteins,” Dr. Schürer said.
Tags: data science, Department of Molecular and Cellular Pharmacology, Dr. Stephan Schürer, drug discovery, molecular and cellular pharmacology, technology