De novo design of luciferases using deep learning
Problem
Designing entirely new enzymes from scratch has been limited by a lack of suitable protein structures and the complexity of protein sequence-structure relationships.
Why it matters
Custom-designed enzymes could enable new catalytic functions and biomedical tools that natural proteins cannot provide.
What the AI did
A deep-learning 'family-wide hallucination' approach generated large numbers of idealized protein scaffolds with diverse pocket shapes, which were used to design artificial luciferases.
Results
Designed luciferases with high selectivity for the luciferin substrates diphenylterazine and 2-deoxycoelenterazine (abstract truncated before full metrics)
Impact
Published in Nature with 441 citations, indicating substantial scientific recognition.
Record
| Venue | Nature |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Andy Hsien-Wei Yeh, Christoffer Norn, Yakov Kipnis, Doug Tischer, Samuel J. Pellock, Declan Evans, Pengchen Ma, Gyu Rie Lee et al. |
| Citations | 441 (OpenAlex, at import time) |
| References | 51 |
| Model or method | deep-learning 'family-wide hallucination' approach |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41586-023-05696-3 |
| Open access | hybrid |
| Keywords | deep learning, protein design, luciferase, enzyme, chemiluminescence |
Importance rating
significant — thousands to millions affected
Abstract calls it a key milestone for computational enzyme design, creating highly active specific biocatalysts from scratch with broad biomedical applications; a genuinely new capability for a foundational field.