ESM Atlas contains more than one billion predicted protein structures and billions more protein sequences. The database was unveiled by researchers at Chan Zuckerberg Initiative Biohub in San Francisco. It exceeds the AlphaFold Database by over 800 million entries and a prior ESM Atlas by about 300 million. Predictions were produced with ESMFold2, which Biohub says outperforms AlphaFold3 and other protein-structure prediction AIs. ESMFold2 is based on a protein language model trained on billions of proteins and includes metagenomic sequences from soil and ocean that are not present in AlphaFold’s predicted-structure database.
Read at www.nature.com
Unable to calculate read time
Collection
[
|
...
]