De novo design of protein structure and function with RFdiffusion
Problem
Designing new proteins with specified structures and functions (binders, symmetric assemblies, enzyme scaffolds) lacked a general deep-learning framework.
Why it matters
Designed proteins can enable new therapies, enzymes and materials that do not exist in nature.
What the AI did
A diffusion model was built by fine-tuning the RoseTTAFold structure prediction network on protein structure denoising tasks, generating protein backbones for diverse design challenges.
Results
not stated in source (abstract gives only qualitative 'outstanding performance' on five design tasks)
Impact
Peer-reviewed in Nature with 2,289 citations, indicating wide adoption of the framework.
Record
| Venue | Nature |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Joseph L. Watson, David Juergens, Nathaniel R. Bennett, Brian L. Trippe, Jason Yim, Helen E. Eisenach, Woody Ahern, Andrew J. Borst et al. |
| Citations | 2,294 (OpenAlex, at import time) |
| References | 70 |
| Model or method | RFdiffusion (fine-tuned RoseTTAFold diffusion model) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41586-023-06415-8 |
| Open access | hybrid |
| Keywords | protein design, diffusion models, RFdiffusion, RoseTTAFold, deep learning, generative modelling |
Importance rating
major — millions of lives, an industry or a field
Abstract states a general deep-learning framework for protein design (binders, symmetric assemblies, enzyme scaffolds, therapeutic and metal-binding proteins) was lacking; hundreds of designs experimentally characterized.