Overview
- Biohub publicly released its fourth‑generation Evolutionary Scale Modeling suite and an ESM Atlas on Wednesday to make large‑scale protein prediction and design openly available.
- The core system uses ESMC, a protein language model trained on billions of sequences, and ESMFold2, which translates model representations into atomic‑resolution structures for design tasks.
- Biohub reports laboratory validation of AI‑designed binders against cancer and immune targets, including PD‑L1 molecules that restored T‑cell signaling in cell assays, though those results appear in a preprint and need peer review.
- The models and the ESM Atlas are available under an MIT license and distributed through biohub.ai and partner platforms such as AWS Bio Discovery and SandboxAQ with compute credits for researchers.
- The release is part of the Chan‑Zuckerberg funded Virtual Biology Initiative, a multi‑year push that includes a reported $500 million commitment to build predictive biology tools and datasets.