Zuckerberg, Chan's Biohub launches protein 'world model'

Non-profit research organisation Biohub has released what it claims is an AI-powered "world model" of protein biology that can drive the prediction, design, and discovery of new therapies.

The philanthropic organisation, set up by Meta chief executive Mark Zuckerberg and his wife Priscilla Chan, said the model will allow mapping of protein patterns in much shorter times, predict their structures, and design new molecules that can bind to them and affect their function.

The world model provides an open discovery engine for protein structure prediction, design, and biological discovery that is being made freely available to researchers around the world. It is based on Biohub's ESM atlas of 6.8 billion proteins and 1.1 billion structures, the ESMC language model that virtually represents proteins, and the ESMFold2 design engine that predicts the 3D structures of proteins and their interactions with binders.

"Proteins…are among the most important targets in medicine, yet designing functional, stable proteins that work as intended in the body is an immense scientific and technical challenge," said Biohub in a statement.

ESMfold2 an alternative to other AI-powered systems like Google DeepMind's AlphaFold and Recursion's Boltz series, and like its rivals promises to reduce the discovery time for new protein binders from months or even years with conventional techniques to days or just a few hours in some cases.

"Importantly, we've built a general model that's capable of an exceptionally broad range of applications across protein biology," said Lori Goler, president of the Chan Zuckerberg Initiative, which supports Biohub.

"We did not train a specific model to design protein binders. We trained a model to understand proteins, and from this, the ability to design protein binders emerged, along with many capabilities," she added. "By learning the underlying rules of how proteins fold, bind, and function, the model can reason about proteins it has never seen – and generate new ones that work."

Biohub researchers have already used the model to design protein binders against five important targets in cancer and immunology – EGFR, PDGFRβ, PD-L1, CTLA-4, and CD45 – achieving hit rates of 36-88% for 'minibinder' proteins and 15-29% for antibody modalities.

"Biohub was built on the belief that open science accelerates discovery," said Chan, who is co-founder of Biohub.

"Making these tools freely available means researchers everywhere can move faster toward personalised cures that work for individual patients, because they target the specific biology driving their disease."