Chan Zuckerberg Biohub releases open ‘world model’ of protein biology

Biohub released ESMC (a large protein language model trained on 2.8 billion sequences), ESMFold2 (state-of-the-art structure prediction and binder design engine), and ESM Atlas (mapping 6.8 billion sequences and 1.1 billion structures). The models successfully designed functional therapeutic protein binders in lab tests with high hit rates. This represents a major open-source advancement in AI-driven biological discovery and therapeutic design.