MIT researchers have developed AbMap, a computational model that predicts antibody structures with improved accuracy. This could accelerate the development of treatments for infectious diseases like SARS-CoV-2. Unlike existing AI models, AbMap focuses on antibodies’ hypervariable regions — highly diverse segments critical for binding pathogens — and overcomes limitations in previous protein modeling approaches.
A new computational technique enables large language models to predict antibody structures with greater accuracy. (Credits: Image: MIT News; iStock)
Using training data from thousands of antibody structures and binding studies, AbMap can evaluate millions of antibody variants to identify the most promising candidates. In collaboration with Sanofi, a global healthcare company, experimental tests confirmed that 82% of the model’s predictions exhibited stronger binding than initial candidates.
Beyond drug discovery, AbMap enables structural comparisons of antibody repertoires, offering insights into individual immune responses. This could help explain phenomena like “super-responders” to diseases such as HIV.
Bonnie Berger, the Simons Professor of Mathematics and head of the Computation and Biology group at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), along with Bryan Bryson, an associate professor of biological engineering at MIT and a member of the Ragon Institute of MGH, MIT, and Harvard, are the senior authors of a paper published in the Proceedings of the National Academy of Sciences. Rohit Singh, a former CSAIL research scientist who is now an assistant professor of biostatistics, bioinformatics, and cell biology at Duke University, and Chiho Im ’22 are the lead authors of this paper. Researchers from Sanofi and ETH Zurich contributed to the research as well. The paper highlights how AI can streamline therapeutic development and lower costs by identifying viable candidates early.
