How Mark Zuckerberg’s Wife & Doctor Founded CZ BioHub & Its Breakthrough ESMfold2 Tool

The Chan Zuckerberg Biohub (CZ Biohub) has announced the release of ESM Atlas, a comprehensive database of protein structure predictions powered by its new AI model, ESMFold2. The research institution claims that ESMFold2 delivers predictive performance that surpasses other existing models, including Google DeepMind’s AlphaFold3.

According to a report in the journal Nature published May 27, 2026, the ESM Atlas represents a significant increase in the scale of available protein data. ESMFold2 has predicted 1.1 billion protein structures, a figure that is 800 million greater than the number of structures contained within the AlphaFold Database.

Technical Framework of ESMFold2

The development of ESMFold2 is based on a protein language model (PLM) that the research team originally released in 2024. A protein language model is designed to treat the amino acid sequences that make up proteins as if they were a language, learning the complex patterns and rules governing how these sequences fold into three-dimensional structures.

To train the model, the researchers utilized billions of protein data points sourced from a wide variety of living organisms. This training allows the AI to predict the 3D shape of proteins based on their linear sequences of amino acids.

Scientific Implications and Utility

Alex Reeves, the Science Chief at CZ Biohub, indicated that the ESM Atlas provides visibility into biological areas that were previously inaccessible to researchers.

ESM Atlas shows even the most unknown protein regions

Alex Reeves, Science Chief at CZ Biohub

Reeves further stated that the database will be a powerful foundation for new biological discoveries.

Institutional Context and Industry Competition

CZ Biohub was established by Priscilla Chan, a physician and the wife of Meta founder Mark Zuckerberg. The institution focuses on leveraging technology to advance biomedical research.

The emergence of ESMFold2 creates a new competitive benchmark in the field of structural biology, which has been heavily influenced by Google DeepMind. DeepMind’s AlphaFold model was recognized with the 2024 Nobel Prize in Chemistry for its transformative impact on the ability to predict protein structures.

While CZ Biohub asserts that ESMFold2 outperforms AlphaFold3 in predictive capabilities and data volume, the report notes that opinions remain divided among some experts regarding the exact performance claims of the new model.