Biohub: The Future of Biology is Open-Source with Co-Founders Mark Zuckerberg, Priscilla Chan, and Head of Science Alex Rives

What It Covers

Mark Zuckerberg, Priscilla Chan, and Alex Rives discuss the Chan Zuckerberg Biohub's $500 million Virtual Biology Initiative, which combines frontier AI with frontier biology to build hierarchical world models of proteins, cells, and biological systems, releasing all tools as open-source to accelerate scientific progress across the entire research community.

Key Questions Answered

• •Hierarchical Biology Modeling: Build biological AI models from the ground up — proteins first, then cells, then whole systems like the immune system. Skipping layers produces weaker models. Biohub's ESM Fold already predicted structures for over 1.1 billion proteins, establishing the protein-level foundation before tackling cellular complexity at the next tier.
• •Frontier Biology as Data Strategy: Unlike language models that train on existing internet text, biology models require inventing entirely new scientific methods to generate training data. Biohub's Chicago hub engineers inflammation-sensing devices, New York develops cellular engineering techniques, and San Francisco runs cryo-EM imaging — each producing novel datasets unavailable anywhere else.
• •Protein Design as Emergent Capability: Rather than building task-specific models, train one general protein language model and gain design capabilities as an emergent property. ESM Fold, trained only to understand proteins broadly, produced nanomolar-binding single-chain antibodies — therapeutically relevant potency — validated experimentally in a 96-well plate screen without any antibody-specific training objective.
• •Predicting Drug Toxicity via Single-Cell Atlas: A comprehensive single-cell transcriptomic atlas can identify off-target drug effects before human trials by revealing which unexpected cell types express a target receptor. If kidney cells express a receptor assumed to be liver-specific, the model flags renal toxicity risk digitally, potentially eliminating a major cause of late-stage clinical trial failures.
• •Open-Source Accelerates Rare Disease Progress: Releasing biology models as open-source tools enables self-organized rare disease patient communities — who already run their own registries, biobanks, and trials — to access capabilities previously unavailable to them. Distributing tools broadly across a long tail of diseases generates scientific knowledge that feeds back into understanding common disease mechanisms.