In brief: The ESM Atlas provides a large-scale map of protein sequences and predicted structures. It is an exploration resource, not an experimental catalogue of validated protein functions.
A new map of protein space
Biohub describes an atlas encompassing 6.8 billion protein sequences and more than one billion predicted structures. These figures refer to datasets assembled or predicted computationally, not to proteins individually validated in a laboratory.
How ESMC and ESMFold2 contribute
ESMC provides learned protein representations that organize related biological patterns. ESMFold2 supplies structural predictions for large subsets of sequence space. The atlas also uses interpretability methods to expose latent relationships.
Practical use cases
Researchers may use the atlas to explore sequence neighborhoods, compare families, inspect predicted structural relationships and generate hypotheses for further work. Correct taxonomic labeling, family annotations and assay design remain independent scientific tasks.
Data scale and provenance
Full datasets can require terabytes of storage and careful data management. Before downloading, review dataset licensing, update schedules and representative subsets. Beginning with cluster-level summaries is often more manageable than mirroring the full corpus.
Primary sources and further reading
Editorial review: October 2026. This guide is educational. Model availability and APIs may change; confirm exact versions in the linked documentation. Generated structures and sequences require independent scientific validation.