Using the ESMFold model, researchers were able to digitally design and then experimentally validate nanomolar binders, a level of activity relevant for therapeutics.

The Chan Zuckerberg Initiative's structural biology center uses cryo-EM microscopes to biophysically and functionally characterize molecules designed by its AI models.

The ESMFold model is achieving state-of-the-art performance across structure prediction benchmarks, particularly for protein-protein and protein-antibody interactions.

Protein language models trained by the Chan Zuckerberg Initiative can learn biological structure and function as an emergent property from token prediction training tasks on protein sequences.

The ESMFold model has folded over 1.1 billion proteins and predicted their structures.

The next major research challenge for the Chan Zuckerberg Initiative is to build a "virtual cell" model that connects the proteomic, genetic, and transcriptomic layers to phenotype.

The Chan Zuckerberg Initiative's new ESMFold is an open system and world model for protein biology based on a language model trained on billions of protein sequences.