Biohub Releases ESMC, a Protein World Model Trained on 2.8B Sequences
Biohub open-sources ESMC, ESMFold2, and a 6.8B-protein Atlas — beating AlphaFold 3 on antibody-antigen prediction and designing lab-validated cancer binders in days
- Biohub open-sourced ESMC, ESMFold2, and ESM Atlas under the MIT license — free for commercial and non-commercial use.
- ESMFold2 outperforms AlphaFold 3 on antibody-antigen complex prediction using ESMC embeddings alone, without MSA input.
- Lab-validated binder design against 5 cancer/immunology targets achieved 36–88% hit rates for minibinders and 15–29% for antibody-derived formats with nanomolar affinity.
- ESM Atlas maps 6.8 billion proteins and 1.1 billion predicted structures — the largest AI application to protein biology to date.
- ESMC trained on 2.8 billion sequences (up from 50M for ESM2) shows log-linear scaling improvements in structural understanding.
- Available now via Biohub Platform (API) or locally via Hugging Face; binder design notebook included.
Biohub has released what it calls a world model of protein biology: three open artifacts , ESMC, ESMFold2, and ESM Atlas , that together form a scientific engine for predicting protein structure, designing new binders, and navigating the protein universe at a scale not previously possible. The release comes from the team behind the ESM model family, which Biohub absorbed when it recruited EvolutionaryScale just seven months ago.
Three pieces, one system
The release bundles three distinct but tightly coupled components:
- ESMC , a protein language model (PLM) trained on approximately 2.8 billion protein sequences drawn from across all of life, including organisms in extreme environments and the 20,000+ protein types in the human body. A PLM works like a text language model but over amino acid sequences: it learns to predict masked residues and, in doing so, internalizes the physical rules governing how proteins fold and function.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
