Biohub launches AI protein model to speed drug discovery

Biohub Unveils AI World Model of Protein Biology for Drug Discovery

Biohub, the philanthropic biomedical research organization backed by Meta CEO Mark Zuckerberg and Priscilla Chan, on Wednesday released what it calls a “world model of protein biology” — a suite of open-source AI tools designed to compress drug discovery timelines from years to hours or days.yahoo

Fourth-Generation Models Map Billions of Proteins

The system is built on the fourth generation of evolutionary scale modeling, or ESM, which learns from the protein sequences produced by evolution to understand protein biology, according to Reuters. The release includes three core components: ESMC, a protein language model trained on billions of protein sequences; ESMFold2, a structure prediction and design model; and ESM Atlas, the largest atlas of protein structure and function covering 6.8 billion proteins across the breadth of life’s biodiversity.biohub

Biohub researchers used the AI models to computationally design new protein binders targeting cancer and immune responses, which successfully reactivated immune cells in laboratory experiments. The models will also be available on platforms including AWS Bio Discovery and SandboxAQ.reuters

Building on the Virtual Biology Initiative

The release builds on momentum from Biohub’s $500 million Virtual Biology Initiative, announced on April 29, which brought together the Broad Institute of MIT and Harvard, the Allen Institute, Arc Institute, Wellcome Sanger Institute, Nvidia, and international consortia including the Human Cell Atlas and Human Protein Atlas. Of that commitment, $400 million is earmarked for internal technology development including next-generation imaging and data infrastructure, while $100 million funds external research.edtechinnovationhub

All data and models generated through the initiative are being made openly and freely available to the global scientific community.biohub

From Meta Research to Philanthropic Mission

The ESM lineage traces back to Meta’s Fundamental AI Research team, which in 2022 released ESMFold and the original ESM Metagenomic Atlas containing 617 million predicted structures. EvolutionaryScale AI, the company that continued developing ESM models, was acquired by Biohub in late 2025. The latest release represents a substantial expansion — from hundreds of millions to billions of mapped protein sequences — and a shift in purpose from research demonstration to practical drug discovery infrastructure.meta