AI for Science: Ada Fang "From Learning Molecular Interactions to AI Agents for Scientific Discovery at Scale"

Abstract:
How can we build AI systems that can both understand and actively explore biology? This talk presents two directions for building more powerful AI systems for scientific discovery. Molecular interactions drive biology. I will first share ATOMICA, which learns representations centered on molecular interactions themselves, rather than individual molecules. These representations are learned using geometric and multimodal approaches at the atomic scale, and the approach leads to a more transferable model across proteins, nucleic acids, and small molecules. The representations not only improve predictive performance, but also enable comparison and reasoning across molecular modalities. The second direction addresses scale and asks how to move from better models to faster discovery with AI scientists. The talk features a multi-agent framework for long-running scientific experimentation that enables AI systems to iteratively propose, critique, and run experiments in a closed loop of hypothesis generation and validation. These directions point toward AI systems that not only model biology, but actively help discover it.

Bio:
Ada Fang is a PhD student in the Department of Chemistry and Chemical Biology at Harvard University, advised by Associate Professor Marinka Zitnik in the Department of Biomedical Informatics at Harvard Medical School. She is also a Research Fellow at the Kempner Institute for the Study of Natural and Artificial Intelligence. Her research focuses on developing AI models to better capture molecular interactions and on building AI systems that enable scientific discovery at scale. Ada has held research positions at Prescient Design, Genentech, and Google DeepMind. She completed her Bachelor of Science at the University of Sydney, Australia.