🔬ESMFold2: The Bitter Lesson is Coming for Proteins - Alex Rives, BioHub
ESMFold2 achieves SOTA protein interaction prediction with inference-time scaling validated across cancer targets
BioHub (Chan Zuckerberg Initiative) released ESMFold2, a protein structure prediction model built on Cryo-EM data that achieves state-of-the-art performance on protein interactions, particularly antibodies critical for therapeutics. The release demonstrates inference-time scaling working across five cancer and immunology targets, extending a key AI scaling law to computational biology. The team also released an atlas of 6.8 billion proteins and 1.1 billion predicted structures as an open scientific resource.