What Happened
On May 27, 2026, Biohub announced a protein-biology platform combining ESMC representations, ESMFold2 structure prediction and design, and an expanded ESM Atlas. The atlas covers 6.8 billion protein sequences, including 1.1 billion predicted structures.
Why It Matters
The release connects large-scale protein representation learning with searchable biological resources and experimental design workflows. Its scale makes previously uncharacterized proteins more accessible for researchers seeking functional relationships and candidates for laboratory investigation.
Technical Details
Biohub describes ESMC as trained on approximately 2.8 billion sequences. ESMFold2 converts sequence representations into structural predictions for proteins and biomolecular complexes. The authors report laboratory validation of designed binders against five oncology and immunology targets. These results demonstrate a research workflow, not clinical effectiveness. Atlas structures remain computational predictions, and successful binding experiments do not establish that a designed protein will be safe or effective as a medicine.