In high school, Jacob Gardner interned for a company that attempted to predict hurricanes. “That was my first exposure to machine learning,” he recalls.
Machine learning, a form of artificial intelligence (AI), works by predicting the future using vast amounts of data from the past. “We train models that take the data—examples we’ve seen—and we try to generalize new examples we haven’t seen,” says Gardner, assistant professor in computer and information science in the School of Engineering and Applied Science.
Today, Gardner applies machine learning not to weather prediction, but to scientific research. Rather than predict the movement of hurricanes, he develops tools that will allow scientists to supercharge fields like drug discovery. “I want to build the AI equivalents of the electron microscope,” he says. “Tools that help scientists do what they’re already doing, just faster, more effectively and with new insight.”
In 2022, Gardner received a CAREER Award from the National Science Foundation to support his research on applying AI to science.
Given the collaborative nature of research at Penn Engineering, Gardner can now employ those techniques in projects with colleagues at the Perelman School of Medicine. “If you want to go from the conception of a drug using AI technology through to clinical trials,” Gardner points out, “Penn actually does clinical trials in house—talk about a collaborative environment.”
Ultimately, Gardner hopes to create the tools necessary to run a fully automated lab, which could discover the next generation of therapeutics for drug-resistant bacteria and diseases like cancer. “The dream is a ‘self-driving lab,’” Gardner says. “You could automate at least the first part of drug discovery, which is identifying interesting molecules.”
In his vision, such a lab would use AI to generate molecular structures for a particular class of therapeutics. Robots would then synthesize the compounds and test the molecules against threats like bacteria, viruses, or cancer cells. The system would then ingest the results of the experiments. “Either it works well,” he says, “in which case your model predicted a great new candidate, or you’ve learned something, and you can update the model.”
Read more at Penn Engineering.
