Marinka Zitnik, PhD, associate professor of biomedical informatics at HMS and associate faculty at the Kempner Institute for the Study of Natural and Artificial Intelligence at Harvard University, has developed PDGrapher—an AI model that identifies treatments capable of reverting diseased cells to a healthy state. Unlike traditional drug discovery approaches that test one protein target at a time, PDGrapher maps the relationships among genes, proteins, and signaling pathways to identify the combinations most likely to reverse disease. In tests across 19 datasets spanning 11 cancer types, PDGrapher ranked correct therapeutic targets up to 35 percent better than comparable models and delivered results up to 25 times faster.

In April 2026, a study led by Arjun Manrai, AB ’08, PhD, assistant professor of biomedical informatics at HMS, and published in Science, found that a large language model outperformed physicians across a range of clinical reasoning tasks, including decisions based on real emergency department patient data. The results make the case that medical AI is ready for prospective clinical trials in real-world care settings. “We tested the AI model against virtually every benchmark, and it eclipsed both prior models and our physician baselines,” Manrai said.