Robust deep learning–based protein sequence design using ProteinMPNN
Problem
Designing amino acid sequences that fold into target protein structures relied on physically based approaches such as Rosetta rather than deep learning.
Why it matters
De novo protein design enables new enzymes, therapeutics and materials for medicine and industry.
What the AI did
ProteinMPNN, a deep learning sequence design method, couples amino acid positions across single or multiple chains and produces sequences that more robustly encode their structures.
Results
52.4% sequence recovery on native protein backbones versus 32.9% for Rosetta; utility and accuracy demonstrated via X-ray crystallography, cryoEM and functional studies.
Impact
Peer-reviewed in Science with high citations and experimental validation of designed sequences.
Record
| Venue | Science |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Justas Dauparas, Ivan V. Anishchenko, Nathaniel R. Bennett, Hua Bai, Robert J. Ragotte, Lukas F. Milles, Basile I. M. Wicky, Alexis Courbet et al. |
| Citations | 2,049 (OpenAlex, at import time) |
| References | 40 |
| Model or method | ProteinMPNN |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1126/science.add2187 |
| Open access | green |
| Keywords | ProteinMPNN, protein sequence design, sequence recovery, deep learning, cryoEM, X-ray crystallography |
Importance rating
major — millions of lives, an industry or a field
De novo protein design underpins enzymes, therapeutics and materials; method applies to a wide range of design challenges with experimental validation, but abstract states no scale numbers.