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

Summary

Mark Zuckerberg, Priscilla Chan, and Alex Rives frame Biohub as a nonprofit built to give the scientific community open-source tools rather than to become a conventional biotech company. They argue that biology is bottlenecked by data scarcity, so progress requires generating new experimental datasets and linking protein, cellular, and system-level models hierarchically. The centerpiece announcement is ESMFold2, described as an open protein-biology world model trained on billions of sequences, capable of rapid atomic-resolution structure prediction and strong performance on protein-protein and protein-antibody tasks. The team says the model has already enabled digital design and lab validation of proteins and single-chain antibodies, including nanomolar binders. Looking ahead, they emphasize the virtual cell, toxicity prediction, and patient-specific intervention design as the next major frontier, with clinical translation likely requiring new research and trial workflows.

Notes

• Biohub’s core aim is to build open-source tools for science, not to operate as a closed commercial platform.
• Chan’s framing is mechanistic: understand an individual’s biology, risk, and intervention points at the person level.
• The original ambition to cure, prevent, and manage all disease is now described as potentially too conservative.
• Biohub has expanded beyond San Francisco into New York and Chicago.
• The virtual biology initiative is meant to span proteins, cells, and whole biological systems.
• Biology is a data-generation problem as much as a modeling problem, because many of the needed datasets do not already exist.
• Biohub’s imaging work is intended to create new observability into previously unseen biological states.
• The modeling strategy is hierarchical: proteins first, then cells, then systems like the immune system.
• Priscilla Chan says the AI and wet-lab groups were intentionally organized as one integrated effort.
• The experimental stack includes transcriptomics, zebrafish development, and cell-cell communication sensors.
• Zuckerberg says Biohub’s open-source posture is deeply tied to empowering individuals, but it must be balanced with biosafety.
• Zuckerberg rejects a centralized superintelligence vision for science.
• Biohub claims a distinctive niche by combining frontier AI with frontier biology in one organization.
• Alex Rives describes Biohub’s mission as taking care of disease.
• Rives argues scientists are motivated by the sense that this mission may finally be achievable.
• Sarah Guo summarizes the research hiring bar as data, tools, compute, talent, and mission.
• Zuckerberg says AI progress can be made by a small but very strong team.
• ESMFold2 is described as an open system for scientific discovery in protein biology.
• The model is trained on billions of protein sequences and produces atomic-resolution structure predictions quickly.
• Biohub says it has already folded over 1.1 billion proteins and connected them through mechanistic interpretability.
• The model is reported to be state-of-the-art on structure prediction benchmarks, especially for protein-protein and protein-antibody interactions.
• Digital design is already producing proteins and single-chain antibodies, with lab tests yielding nanomolar binders.
• The workflow narrows thousands of digital trajectories into a small wet-lab batch for validation.
• Priscilla Chan emphasizes lab validation with antibody screens, cell assays, cryo-EM, and structural microscopy.
• The larger medical promise is to predict off-target effects before human testing.
• Chan points to renal toxicity as an example of an off-target effect discovered too late in development.
• Zuckerberg says rare-disease patient groups can self-organize to improve trial recruitment economics.
• The team sees clinical translation as dependent on changing how clinical research is done.
• Zuckerberg says Biohub became the main focus of their philanthropy over the last year.
• The leadership model has shifted from biologists interested in technology to an AI researcher with biology background.
• Biohub is aiming for hierarchical world models around biology and world-class output over the next five years.
• The next major frontier is the virtual cell, a model that ladders up biological complexity to cell-level behavior.
• The virtual cell should integrate proteomic, genetic, and transcriptomic layers and connect them to phenotype.
• Generalization to unseen interventions will require an enormous data-generation effort.
• Rives says agentic systems are already being coupled to biology models for automated design.
• Zuckerberg says Biohub is compute constrained and must balance protein and cellular work.
• Model size is constrained by the scale of the data available.
• Biohub’s strategy is to build hierarchical world models and do the highest-quality work in biology.
• Biohub’s main strategic change over the last year is that philanthropy is now focused there.
• Chan says the teams are now closely interlocked rather than merely aligned.
• Chan says this tighter integration is exciting, but also a little scary.