Open-source AI model ESMFold2 predicts structures of 1 billion proteins (nature.com)

• ESMFold2 predicts over 1 billion protein structures.
• Atlas includes 6.8 billion protein sequences, mostly metagenomic.
• Model is fully open-source and surpasses AlphaFold3.
• Successful design of new antibodies validated in lab.

"The Chan Zuckerberg Biohub released the ESM Atlas, an open-source database of over 1 billion predicted protein structures generated by the AI model ESMFold2. It surpasses the AlphaFold database by more than 800 million entries and claims to outperform AlphaFold3, especially on antibody-protein complexes. The model is fully open-source and trained on billions of metagenomic sequences."