Chan Zuckerberg Initiative’s Biohub unveiled an AI-generated protein atlas, the ESM Atlas, expanding predicted protein structure coverage by more than a billion entries. Biohub said the database, built with its ESMFold2 model, surpasses its prior ESM Atlas iteration and eclipses the AlphaFold Database in total predicted structures. Biohub positioned ESMFold2 as better at modeling protein complex structures, including antibody–antigen interactions, and said the approach is based on a protein “language” model trained on billions of proteins. The team also highlighted that ESMFold2 is fully open source, aiming to accelerate antibody and protein engineering workflows. The release adds to an increasingly competitive field of open and proprietary protein modeling systems, with the number of predicted structures and complex modeling accuracy now emerging as key differentiators for drug-discovery use cases.