It can take 10-15 years from the time a drug candidate is discovered to the time it finally hits the market, all with a price tag averaging around $2.6 billion. And this is only the best case scenario, one where the right drug formulation was established from the very beginning.
Of all the drugs invented in a lab, few make it past a safety and toxicity study in humans, which is only the first phase of three needed to pass a clinical trial. Data suggest that as many as 88% of new drugs fail at this early stage of the process.
In recent years, artificial intelligence (AI) has emerged as a tool that has the potential to accelerate the drug discovery process. A quickly emerging partner in this mission is Watson, a first-in-class supercomputer developed by IBM Watson Health. It is capable of complex computing and is commercially available to everyone, just like an app on a smartphone.
On top of that, Watson can understand complicated questions posed in natural human language and derive meaning from them, proposing evidence-based answers. And much like a human brain, its knowledge improves over time from experience.
Watson has been well-known within the AI field since 2011, but IBM only recently began to explore different applications, finally adapting it for medical research and drug discovery. By taking advantage of a computer system that can process millions of documents faster than any human could, research scientist Dr. Naomi Visanji at Toronto Western Hospital became the first to apply Watson for research in Parkinson’s disease.
IBM Business Development Executive and Parkinson’s patient Jonathan Rezek and Toronto Western Hospital research scientist Naomi Visanji talk about their collaboration to use IBM Watson Health to find new ways to treat Parkinson’s disease.
Parkinson’s patients are only diagnosed 20 years after the onset of the disease when symptoms, like worsening motor control, become obvious. Currently, standard treatment includes levodopa, a drug that restores levels of the neurotransmitter dopamine in the brain to improve motor control. Unfortunately, as the disease progresses, levodopa becomes less and less effective until there are no more healthy neurons left to absorb the dopamine from the drug.
The idea here is that there might already be an approved drug that can treat Parkinson’s, but we might not know it yet. It might have been created for another purpose, but if it’s in the literature, we might already know a lot about it.
Under Visanji’s guidance, Watson searched the entire scientific database (more than 20 million summaries of scientific studies available online) to try to find that drug. Researchers trained Watson to look for keywords such as alpha-synuclein, a brain protein thought to mediate Parkinson’s, while searching for an already approved drug in Ontario.
Watson’s search resulted in a ranked list of 52 drug candidates. Visanji then identified 21 of those worthy of further study in the lab, including 16 that had never been linked to Parkinson’s before. Now her job is to figure out which are the best candidates in the lab and their effects on alpha-synuclein aggregation, the main molecular hallmark of Parkinson’s.
Using artificial intelligence for drug screening may provide a shortcut to reduce the years of basic science research, and even perhaps the toxicology tests needed to get an existing drug approved for a new disease. Knowing that it has already passed safety and toxicity tests in a previous clinical trial, researchers can move ahead with more confidence that it can be repurposed.
If Watson succeeds in its mission and finds an existing drug that could treat Parkinson’s, it will mark a new era of medical research: one where the so-called Valley of Death that consumes so many potential treatments before they can make it to market allows a few more to pass.