How IBM’s Supercomputer Found a Hidden New Parkinson’s Drug

See? Bots aren’t all bad.

Artificial intelligence has been getting a bit of a bad rap recently—albeit, for good reason. The launch of OpenAI’s ChatGPT has ignited fears that it’ll take over writing jobs of all types. Meanwhile, headlines claimed that Bing’s new AI-powered search engine has been doing everything from threatening to falling in love with users.

However, there are still applications of AI that undoubtedly do a lot of good for the world. Case in point: Researchers used an AI to discover a potential new treatment for Parkinson’s disease in a new study published on March 2 in the journal PLOS Biology. The bot analyzed the language in scientific literature for drugs unrelated to Parkinson’s in order to find promising drug candidates.

To understand how the AI found this drug, you need to first understand something called mitochondrial dysfunction. This occurs when the powerhouse of a cell (i.e. the mitochondria) doesn't work properly. It’s further exacerbated when the cells can’t dispose of the defective mitochondria in a process called mitophagy.

Mitochondrial dysfunction can be caused by a variety of diseases and conditions. However, it’s also a contributor to neurodegenerative issues like Parkinson’s, dementia, and Alzheimer’s. That’s why researchers want to find new compounds that might promote mitophagy in the human body.

But there’s a problem: Many of the compounds that support mitophagy are also harmful to the cells—which make them bad candidates for treatments. However, the researchers believed that the language used in the published academic literature to describe these compounds might actually lead to other drugs that haven’t previously been linked to mitophagy.

“Using an artificial intelligence platform, we employed a natural language processing approach to evaluate the semantic similarity of candidate molecules to a set of well-established mitophagy enhancers,” the study’s authors wrote.

Specifically, the team utilized IBM’s Watson for Drug Discovery, a powerful AI that uses IBM’s famed Jeopardy-playing supercomputer to crawl through and analyze scientific articles and journals for specific keywords and phrases. The team was able to develop a semantic “fingerprint” for compounds that support mitophagy. After finding 79 candidates, they tested the compounds to see if they promoted the process and identified one: probucol, a drug used to lower cholesterol.

The researchers found that the drug was not only effective at promoting mitophagy, but it also improved motor function and survival of neurons in two different animal models with Parkinson’s disease (zebrafish and fruit flies).

“In our study, we used the AI-platform IBM Watson to efficiently identify currently approved drugs that could potentially be re-purposed as therapies for Parkinson’s disease,” Angus McQuibban, a biochemist at the University of Toronto and lead author of the study, said in a statement.

While this is undoubtedly good, it should be noted that IBM Watson doesn’t have a perfect track record when it comes to giving medical advice.

In 2018, the company drew criticism after the supercomputer gave “unsafe and incorrect treatment recommendations” to hospitals and doctors caring for cancer patients. So even with this potential new Parkinson’s treatment, it’s important to remember that humans should always be in the loop when it comes to AI.

Of course, more research is needed to ensure that the probucol is an effective treatment for Parkinson’s in humans. However, this experiment does show AI isn’t all scary chatbots and creepy computer-generated images. When wielded correctly and with plenty of human supervision, it can help bolster scientific discoveries—and potentially treat one of the most destructive diseases out there.

Source