BioHub Folds 1.1 Billion Proteins Without Designing Model for Specific Diseases
"We folded over 1.1 billion proteins and predicted their structures, and we didn't design a model for antibodies. We didn't design a model to be able to bind one particular target. We just designed a model that could understand proteins. You get protein design as an emergent property."
About this episode
On this episode of No Priors, hosts interview Mark Zuckerberg, Priscilla Chan, and Alex Reeves about the Chan Zuckerberg Initiative's BioHub and its ambitious virtual biology initiative, which has now become the couple's primary philanthropic focus with a $500 million commitment. The conversation reveals that when Zuckerberg and Chan first proposed curing all diseases by 2100, Nobel Prize-winning scientists literally laughed at them—until the couple pressed them to explain why, uncovering that the real barriers were organizational rather than scientific. This insight shaped BioHub's strategy of building open-source tools to accelerate the entire scientific field rather than pursuing specific cures. Alex Reeves, who recently joined from evolutionary scale research, detailed how BioHub's new ESMFold model folded 1.1 billion proteins and achieved nanomolar antibody binding in single 96-well plate experiments—compressing what traditionally required screening millions of antibodies into computational design plus minimal lab validation. The discussion emphasized BioHub's unique positioning as the only organization combining frontier AI research with frontier wet-lab biology, generating novel datasets that don't exist elsewhere through cellular engineering, advanced imaging, and inflammation sensors. Zuckerberg argued that current 100-year disease cure timelines are now too conservative given AI progress, and outlined a vision where virtual cell models could enable digital clinical trials, fundamentally disrupting the $1.5 billion, 15-year drug development process. The team explained their hierarchical approach to building world models of biology—starting with proteins, scaling to cells, and eventually to whole systems like the immune system—with all work released as open source to empower the broader scientific community rather than centralizing discovery.
Key takeaways
- BioHub scientists designed therapeutic-level antibodies using AI in a single 96-well plate experiment after computational screening of hundreds of thousands of candidates.
- Nobel Prize winners initially laughed at Zuckerberg and Chan's goal to cure all diseases, calling their century timeline too ambitious—now they say it's too conservative.
- BioHub folded 1.1 billion proteins without designing disease-specific models, with protein design capabilities emerging spontaneously from general-purpose architecture.
- Zuckerberg predicts virtual cell models will enable digital clinical trials within years, potentially disrupting the $1.5 billion, 15-year drug development timeline.
- The organization is the only one combining frontier AI research with frontier wet-lab biology, generating novel datasets through cellular engineering and advanced imaging.
- All BioHub models are released as open source to accelerate the entire scientific field rather than pursuing specific disease cures internally.
- Patient groups organizing their own clinical trials and biobanks have moved gene therapies from concept to treatment in 3-5 years versus decades through traditional paths.