Convergent Research

Interdisciplinary BU Team Teaches AI the Biology of Antibodies, Speeds Drug Discovery

Much of the AI race has focused on scale: bigger models, more data, and more compute. But for specialized scientific problems, what if the better approach is teaching AI what matters rather than simply making it bigger?

New research from Boston University, published in the Nature Portfolio journal Communications AI & Computing, explores that question in antibody discovery. By focusing on the antibody regions most important for recognizing and binding to disease targets, the researchers improved binding-affinity predictions by as much as 27% compared with much larger models, while using fewer computational resources. The findings point to the potential of biologically informed AI to help researchers better understand antibody-antigen recognition, prioritize candidates for laboratory testing, and inform the design of more effective therapies.

The interdisciplinary team behind this work includes John Misasi, MD, assistant professor, virology, immunology & microbiology, who also serves as a National Emerging Infectious Diseases Laboratories core faculty member.

Learn more.