Revolutionizing drug research for Parkinson’s with AI’s Watson

At the age of 50 I was diagnosed with Parkinson’s disease—a chronic, neurodegenerative movement disorder for which there is no cure. Globally, roughly ten million people are affected by Parkinson’s. The mainstay symptomatic therapy, L-DOPA, is more than 50 years old. While L-DOPA is effective at reducing the symptoms of Parkinson’s for a few years, eventually it causes debilitating unwanted movements, or dyskinesia in the majority of patients taking it. After ten years of L-DOPA, up to 90 percent of patients are affected by dyskinesia—in other words, the treatment ends up being worse than the disease.

Facing these facts, the first few years after my diagnosis were bleak. Then finally one day, I had a eureka moment—could we revolutionize how new Parkinson’s treatments are being researched by using artificial intelligence?

Watson drug discovery capabilities

At IBM, I teach startups how to use Watson, IBM’s artificial intelligence platform. I know Watson’s capacity for ingesting and analyzing data, and its ability to reason and learn. One of the key challenges with finding new treatments is that getting a new drug approved for the market is a monumental task at a cost of over $2.5 billion and over a timeframe of approximately 10 years.

An easier approach is to identify drugs already approved to treat one condition, that are safe for human use, and then repurposing them for another condition. Given what I’ve seen with Watson, I thought its drug discovery capabilities could potentially accelerate finding an existing drug that could be used off-label to improve Parkinson’s patients’ quality of life.

I took the idea to my neurologist, Dr. Connie Marras. She was curious and brought in her research colleagues at the University Health Network’s Morton and Gloria Shulman Movement Disorders Centre (MDC) in Toronto, including Dr. Naomi Visanji. We formed a team—IBM Watson Health, the Ontario Brain Institute and the MDC.

Cutting through unchartered territory

First the team put together training information about dyskinesia for Watson and narrowed the scope of the study to 3,500 drugs that met some basic criteria for having the potential to be repurposed for Parkinson’s. Then within roughly 30 minutes, Watson was able to read the relevant scientific literature, absorb patterns and draw parallels between related information. The results were encouraging—within Watson’s top five percent of ranked candidate drugs, some were already known to have some potential to impact dyskinesia in pre-clinical studies. This gave us hope that Watson might find new effective drugs within its top ranked candidates.

Examining the list further, Dr. Naomi Visanji and her colleagues flagged five candidates with novel plausible antidyskinetic mechanisms of action that to her knowledge no one had considered before. According to Dr. Visanji, the first of these five drugs is undergoing lab testing and showing some exciting preliminary results that it’s working against dyskinesia. Just knowing about the other possible drugs however, wasn’t enough; they still had to be clinically tested.

Putting Watson results to the test

As a testament to our work and its potential to revolutionize how Parkinson’s patients are treated, the Canadian Institutes of Health Research (CIHR) awarded our team a grant in January 2018. This grant will cover the costs of testing the remaining top candidates in a well-established rodent model of dyskinesia and performing additional pharmacological studies that will ease the translation of any findings to future studies in primates.

Our application ranked in the top 6% of those received by the CIHR, in what is an extremely competitive environment. Our work to date could potentially impact and improve Parkinson’s patient quality of life in a matter of just a few years, easing the burden of L-DOPA-induced dyskinesia. While not a cure for Parkinson’s per se, it’s something that we hope might help improve patient’s quality of life until we have a cure.

Source: Dr. Nicky Hekster IBM.