CMU-UBC Team Emerges Victorious in Challenge to Advance Parkinson’s Disease Research

Familial Parkinson’s disease affects at least 500,000 Americans, but a team of Carnegie Mellon University and University of British Columbia researchers have found a potential target for a treatment.

“Leucine-rich repeat kinase 2 (LRRK2) is the most common genetic cause of Parkinson’s disease, and the specific domain we were targeting had no known small molecule or natural inhibitors,” said Ben Koby, a graduate student in Carnegie Mellon’s Department of Chemistry(opens in new window). “We found 14 well-performing molecules in two leading series of compounds.”

Koby, along with fellow graduate students Filipp Gusev and Evgeny Gutkin, Associate Professor of Chemistry Olexandr Isayev(opens in new window) and Professor of Chemistry Maria Kurnikova(opens in new window), were part of the Carnegie Mellon team that took on the Critical Assessment of Computational Hit-Finding Experiments (CACHE) Drug Discovery Challenge #1. The challenge involved using novel computational methods to find potential binding agents for the WD40 repeat (WDR) domain of LRRK2, a donut-shaped hole difficult for most molecules to attach to.

“Finding small molecules that can bind specifically to a target protein with a relatively featureless surface is a notoriously difficult task, both experimentally and theoretically,” Kurnikova said. “That is why no known inhibitors of WD40 existed until this challenge.”

The team started out with an extremely large database of commercially available small molecules and ligands. Collaborators at the University of British Columbia used a technique known as deep docking, an artificial intelligence-driven technique that can rapidly and accurately see if small molecules can potentially fit in target binding sites, to narrow that pool from about 4.5 billion molecules to 37 million potential candidates. They further narrowed the pool through consensus docking, where multiple docking programs are used to ensure there is limited bias in the results. Fuqiang Ban, a medicinal chemist at the University of British Columbia, refined the results by pointing out some potentially successful candidates.

After the initial pool had been narrowed down to 800 molecules, the Carnegie Mellon chemists used absolute binding free energy simulations to determine which of the potential molecules would be optimal for experimental validation. Absolute binding free energy simulations are very accurate but require a significant amount of computational power. The team used Bridges-2(opens in new window), the flagship supercomputer of the Pittsburgh Supercomputing Center (PSC), a joint research center of Carnegie Mellon and the University of Pittsburgh. Bridges-2 allowed them to simulate 800 molecules over six weeks. Seventy-six of the simulated molecules were predicted to be capable of binding to LRRK2. Five of these compounds were experimentally found to be active.

“A key thing in this process is the combination of several computational methods,” Gutkin said. “Machine learning and molecular modeling methods combined together can increase prediction power dramatically.”

From the five top molecules, they took two of the most promising candidates and used a machine learning database to select potential analogs and orientations that would be most likely to strongly bind to LRRK2. After each analog was selected, they performed lead optimization, where they again simulated then thoroughly analyzed the candidate to determine its effectiveness.

“This methodology allowed us to find regions in chemical space that were most populated with extremely good molecules. That allowed us to efficiently select potentially great molecules for computational and experimental evaluation at a substantial fraction of the typical cost,” Gusev said.

After submitting the best analogs to be tested experimentally by the CACHE organizers, they discovered that multiple candidates can inhibit LRRK2. With knowledge about these inhibitors, drug development researchers can use them as a basis for potential Parkinson’s disease treatments.

Based on the novelty of their methods and the effective results, the Carnegie Mellon and University of British Columbia team tied for first place in the CACHE Drug Discovery Challenge #1(opens in new window). The other winning team was led by David Koes from the University of Pittsburgh.

“The beauty of our method is in the mutually beneficial combination of two distinct approaches, machine learning and physics-based simulations, that are synergistic in their strengths,” Gusev said.

For the graduate students Gusev, Gutkin and Koby, finding a potential treatment for Parkinson’s disease is just the start. The active work testing potential binding molecules took about three months compared to at least three or four years of traditional drug discovery preliminary testing. Their method also requires fewer physical resources, since there are fewer molecules that need to be tested experimentally. They believe their method could apply to any potential treatment target, saving time and money.

“It’s about $4 billion and approximately 10 to 15 years to bring a drug to market from start to finish,” Gutkin said. “Accelerating the timeline and reducing the cost is required, and computer-aided drug design is a very promising direction of how to achieve this improvement.”

The team plans to further develop their methods and test them on rare, untreatable diseases. Koby will investigate kinases implicated in diseases like diffuse intrinsic pontine glioma, a fatal form of childhood brain cancer, and fibrodysplasia ossificans progressiva, a disease in which a person’s muscles and tendons turn into bone.

“The beauty of our workflow is that it should be generalizable to any specific protein target,” Koby said. “We’re really looking forward to advancing this methodology and targeting proteins important for other diseases.”

Isayev emphasized that the team would not have performed so well in the competition or created such a promising drug discovery framework without Gusev, Gutkin and Koby.

“We assembled a dream team from three academic labs, and we possessed necessary but complementary expertise needed to solve this problem,” Isayev said. “Professor Kurnikova’s and my lab are working together on several drug discovery related projects, and we’re very much looking forward to applying this methodology to other protein targets. The teamwork and personal relationships between Filipp, Evgeny and Ben played a big role. They are the three musketeers of computational chemistry.”

The Carnegie Mellon team including Chamali Narangoda, a postdoctoral researcher in the Kurnikova group, partnered with Francesco Gentile, Fuqiang Ban, and Artem Cherkasov at the University of British Columbia. Their work received funding from the National Institutes of Health and the National Science Foundation.