Research explores potential of AI to repurpose drugs for Parkinson’s disease

Naomi Visanji met Jonathan Rezek in 2015.

Rezek, an IBM executive who had Parkinson’s disease, was a patient in the movement disorders clinic at Krembil Brain Institute, University Health Network.

He had the idea that Visanji, then a scientific associate with Krembil, work with her colleagues and IBM’s Watson artificial intelligence (AI) to look for new treatments for Parkinson’s disease.

Watson is a computer system with natural language processing and evidence-based learning abilities, and capacity to generate hypotheses. It was best known at the time for winning the quiz show Jeopardy.

Intrigued, Visanji began work with a diverse team that included data scientists, pharmacologists and clinicians. They used Watson to look for existing approved drugs that could be repurposed to treat L-DOPA induced dyskinesia (LID), a debilitating and common side effect of dopamine-replacement therapy L-DOPA in Parkinson’s disease.

“The drug development process takes a huge amount of time, from the fundamental discovery of the target to even basic demonstrations of efficacy,” says Tom Johnston, a key part of the research team and COO at neuroscience contract research organization Atuka.

“If you can take existing drugs that have already crossed those hurdles, then you're potentially saving a lot of time and money.”

The researchers trained Watson to ‘read’ more than one million freely available abstracts from scientific papers and identify drugs with potential anti-dyskinetic effects based on language in the abstracts.

An algorithm then ranked the drugs, which included some expected and unexpected candidates.

“It almost felt like having 100 master’s students actually searching the literature and then laying all of the suggestions out and finding connections between them,” says Johnston, though what would take students months or years was done in a matter of minutes with Watson.

The team created a list of top candidates, then tested three of the drugs in validated animal models, with their selection based on the drug having a novel mechanism of action yet to be tested in LID and an ability to penetrate the blood-brain barrier. They found two of the candidates had no effect, while the third robustly reduced LID, albeit at the expense of L-DOPA’s ability to treat Parkinson’s symptoms.

The study, published earlier this year, did not yield a new treatment for LID. But it did provide proof of principle that advanced data processing and AI can help scientists identify new avenues of research.

“LID is such a multifaceted problem and remains one of the biggest challenges in treating Parkinson’s disease,” says Visanji, now a principal investigator at the University of Toronto’s Tanz Centre for Research in Neurodegenerative Diseases and an affiliate scientist at Krembil.

“There are many potential indications for existing drugs, and it was interesting that we hadn’t thought about some of these drugs for LID before. Watson showed us that we shouldn't limit our thinking.”

Visanji says the system did suggest candidates that were obvious, but that the team was most interested in those drugs that had a plausible mechanism of action but hadn’t been tested as treatments for LID before. “Watson came up with some suggestions that at first seemed a little bit out of left field, but on closer consideration still had a rationale,” she says.

The study was among the first to explore the use of AI in drug discovery for Parkinson’s disease. Visanji says it may help other researchers see the value of AI, and natural language processing in particular.

“It opened my eyes and turned me into a bit of a believer,” she says. “There are now many AIs out there, and a growing number of studies to repurpose drugs based not just on natural language processing but increasingly incorporating details like molecular structures. I hope we’ve helped change a few minds.”

Graham Collingridge is director of the Tanz Centre and a professor of physiology at U of T’s Temerty Faculty of Medicine. He says he is delighted to see the application of AI for research into Parkinson's disease.

“The use of AI, where Toronto is at the forefront internationally, will increasingly impact research into neurodegenerative diseases,” Collingridge says. “We need to capitalize on the opportunities that will arise from this work.”

The study was funded by the Ontario Brain Institute, Canadian Institutes of Health Research and the Blidner Family Foundation.