Drug discovery is traditionally a high-risk and resource-intensive process — so much so that it has drawn comparisons to gambling.
Brendan Frey, a U of T professor, put it bluntly: “It’s like the Big Pharma companies come into a casino, put a million-dollar coin into a slot machine, and with some probability like 10 per cent or something, they get a win.”
But recently, a growing trend in the field is reducing uncertainty around drug discovery by using artificial intelligence (AI) as a prediction tool. Dr. Christine Allen, a professor at the Leslie Dan Faculty of Pharmacy, together with post-doctoral researcher Pauric Bannigan, recently published a review paper on the subject in the Journal of Controlled Release.
How AI can be used to reduce risks in drug discovery
“Let’s say that in our lab… we have a drug, and [it’s] really hydrophobic [repellent of water],” said Allen in an interview with The Varsity. To give such a drug orally, which would expose it to water, researchers must decide on the components that they will need to use in order to make the tablet or capsule ready for delivery. They must also decide on the ratio between those components and the active drug.
Researchers normally conduct a high number of experiments to find these solutions, Allen explained. However, the emergence of prediction tools based on AI can significantly change the process of experimentation.
As Bannigan describes, AI prediction tools have the potential to narrow the starting point from which researchers have to begin experimenting from. By eliminating incompatible solutions, AI can guide scientists toward potential avenues for success, thus saving both time and money.
Guided by these tools, pharmaceutical researchers may hypothetically only need 10 experiments to test promising solutions with AI, as opposed to 100 experiments to test most possibilities, he explained.
Case studies of AI used in drug discovery
Real-life examples of the applications of AI were drawn from the book Prediction Machines: The Simple Economics of Artificial Intelligence, which was cited multiple times in the review paper.
Dr. Avi Goldfarb, a professor at the Rotman School of Management, is a co-author of the book. He wrote to The Varsity that AI prediction tools can have significant and specialized meaning to the pharmaceutical industry.
For example, Atomwise, a company that predicts the binding of molecules with proteins, can “increase the success [rate] of early stage experiments in the drug discovery process and increase the number of successful drugs that come to market.”
BenchSci is another company that makes it easier for scientists to search the relevant literature by predicting which content is relevant to a particular need.
“[BenchSci] is also aimed at improving the drug discovery process,” wrote Goldfarb.
There is a growing trend of Big Pharma companies partnering with those specializing in AI, according to the review paper.
As an example, Allen recalled Novartis, which “dealt with Intel to try and reduce the amount of time required to analyze microscopic images.”
Allen’s research group has now started collaborating with U of T professor Alán Aspuru-Guzik, who has significant expertise in applying AI to chemistry. The teams have been working together to use algorithms that could help predict which materials could be best used for drug discovery.
The impact of AI on human involvement
As for the impact of AI on researchers, Allen noted that as AI tools get more involved in industries, human judgement remains highly valued, and is one of the main ideas of Prediction Machines.
“You might predict the likelihood of rain, but without judgment on how much you mind getting wet and how much you mind carrying an umbrella then the prediction alone won’t tell you what to do,” wrote Goldfarb.
While AI could guide researchers by providing predictions, Goldfarb noted that their human judgement would still be valuable in deciding what to do with the predictions once they have them.