Mark Zuckerberg and Priscilla Chan’s Biohub has unveiled a major breakthrough in AI-driven biology with the release of its new Evolutionary Scale Models (ESM) platform — an open system designed to map, predict, and even design proteins at unprecedented scale.
At the center of the announcement is ESMFold2, a next-generation model built on a protein language model called ESMC, trained on an enormous dataset of 2.8 billion protein sequences. The goal is ambitious: give researchers the ability to predict protein structures and engineer entirely new proteins faster and more accurately than ever before.
According to Biohub, ESMFold2 achieves state-of-the-art performance in protein structure prediction, including protein-protein interactions and antibody-antigen modeling — reportedly outperforming systems like DeepMind’s AlphaFold in several benchmarks.
What makes this announcement especially important is that the models are already showing practical laboratory results. Researchers have reportedly used the system to design binders targeting five cancer and immune-related disease pathways, with hit rates ranging from 36% to 88%. In biotechnology and drug discovery, those are highly meaningful early-stage numbers.
Another major component of the release is ESM Atlas, a massive biological mapping system containing:
- 6.8 billion protein sequences
- 1.1 billion predicted protein structures
The atlas helps uncover previously unknown evolutionary relationships between proteins, potentially opening the door to discovering entirely new biological mechanisms and therapeutic pathways.
This is part of Biohub’s broader $500 million “Virtual Biology Initiative,” which aims to build open AI infrastructure for biological research. Instead of limiting advanced drug-discovery tools to a handful of pharmaceutical giants, Biohub is pushing toward democratized scientific infrastructure — putting powerful computational biology capabilities into the hands of researchers worldwide.
The implications are enormous.
Traditional drug discovery is slow, expensive, and heavily dependent on trial-and-error experimentation. AI systems like ESMFold2 shift much of that process into simulation and prediction, dramatically compressing the time needed to identify promising therapeutic candidates.
We are now seeing a convergence of:
- Large-scale biological datasets
- Foundation models trained on evolutionary information
- High-performance compute
- AI-guided protein engineering
Together, these advances are beginning to reshape biotechnology the same way large language models reshaped software and knowledge work.
Alongside efforts like Isomorphic Labs, Biohub’s work moves the industry closer to the long-term vision described by Demis Hassabis — using AI to dramatically reduce, and potentially one day eliminate, many forms of disease.
We are still early in this transition, but the direction is becoming increasingly clear: AI is evolving from a productivity tool into a scientific discovery engine.
https://biohub.ai/esm/protein/about
Add to favorites
