Biohub researchers released an AI-generated protein atlas dubbed the ESM Atlas, predicted to cover more than one billion protein structures along with billions of sequences, vastly expanding on previous public resources. The atlas was generated using an ESMFold2 model, which Biohub says surpasses the performance of AlphaFold3 for complex prediction tasks. The release positions the resource as open and designed to support discovery by highlighting parts of protein biology that remain unknown. Biohub also said ESMFold2 performs especially well in modeling protein complexes, including antibody-antigen interactions. The expanded dataset is likely to influence antibody discovery workflows and translational research—especially as the field increasingly relies on “AI-designed” candidates that still require experimental validation. With multiple protein-structure platforms racing to provide open and proprietary tools, the scale and open-source nature of ESMFold2 may accelerate both internal discovery programs and collaborative research efforts.