Artificial intelligence (AI) is becoming increasingly popular. It powers models that help us drive cars, proofread emails, and even design new molecules for drugs. But just like humans, it’s difficult to read AI’s mind. Explainable AI (XAI), a subset of this technology, could help us do that by providing reasons for a model’s decisions. And now researchers are using XAI not only to take a closer look at predictive AI models, but also to delve deeper into the field of chemistry.

The researchers will present their findings at the fall meeting of the American Chemical Society (ACS). ACS Fall 2024 is a hybrid meeting, held virtually and in person from August 18 to 22, and will include approximately 10,000 presentations on a range of scientific topics.

Because of the wide range of uses of AI, it is nearly ubiquitous in today’s technology landscape. However, many AI models are black boxes, meaning it is not clear what exact steps are taken to produce an outcome. And if that outcome is, for example, a potential drug molecule, not understanding the steps could generate skepticism among scientists and the public alike.

As scientists, we like reasoning. If we can develop models that provide insight into AI decision-making, scientists could potentially become more comfortable with these methods.”

Rebecca Davis, Professor of Chemistry, University of Manitoba

One way to provide that rationale is through XAI. These machine learning algorithms can help us peek behind the scenes of AI decision-making. Although XAI can be used in a variety of contexts, Davis’ research focuses on applying it to drug discovery AI models, such as those used to predict new antibiotic candidates. Given that thousands of candidate molecules can be reviewed and rejected to get just one new drug approved—and antibiotic resistance is a constant threat to the effectiveness of existing drugs—accurate and efficient prediction models are critical. “I want to use XAI to better understand what information we need to teach computers chemistry,” says Hunter Sturm, a chemistry graduate student in Davis’ lab who is presenting the work at the meeting.

The researchers began their work by feeding databases of known drug molecules into an AI model designed to predict whether a compound would have a biological effect. They then used an XAI model developed by Pascal Friederich at the Karlsruhe Institute of Technology to examine the specific parts of the drug molecules that led to the model’s prediction. This helped explain why a particular molecule was active or not according to the model, and it helped Davis and Sturm understand what an AI model might consider important and how it creates categories after examining many different compounds.

The researchers found that XAI can spot things that humans might have missed; it can consider far more variables and data points simultaneously than a human brain. While screening a number of penicillin molecules, for example, XAI found something interesting. “Many chemists think that the core of penicillin is the key site for antibiotic activity,” says Davis. “But that’s not what XAI saw.” Instead, it identified structures attached to that core as the key factor in its classification, not the core itself. “That might be why some penicillin derivatives with that core have low biological activity,” explains Davis.

In addition to identifying important molecular structures, the researchers hope to use XAI to improve predictive AI models. “XAI shows us what computer algorithms define as important for the effect of antibiotics,” explains Sturm. “With this information, we can then train an AI model on what to look for,” adds Davis.

Next, the team will work with a microbiology lab to synthesize and test some of the compounds that the improved AI models predict would work as antibiotics. Ultimately, they hope XAI will help chemists develop better, or perhaps entirely different, antibiotic compounds that could help stem the tide of antibiotic-resistant pathogens.

“AI creates a lot of suspicion and uncertainty among people. But if we can ask AI to explain what it does, the technology is more likely to be accepted,” says Davis.

Sturm adds that he believes AI applications in chemistry and drug discovery represent the future of the field. “Someone has to lay the groundwork. And that’s exactly what I hope to do.”

The research was funded by the University of Manitoba, the Canadian Institutes of Health Research and the Digital Research Alliance of Canada.