Learning deep representations of enzyme thermal adaptation
Problem
The relationship between protein sequences and thermal adaptation is poorly understood, hindering engineering of thermostable enzymes.
Why it matters
Thermostable enzymes matter for basic biology, drug discovery, and industrial protein engineering.
What the AI did
DeepET, a deep neural network trained on over 3 million BRENDA enzymes labeled with source-organism growth temperatures, learns temperature-related protein representations and predicts optimal catalytic and melting temperatures via transfer learning.
Results
DeepET outperforms classical regression models on rationally designed features and other deep-learning representations (no numbers in abstract).
Impact
Abstract states DeepET holds promise for understanding thermal adaptation and guiding thermostable enzyme engineering.
Record
| Venue | Protein Science |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Gang Li, Filip Buric, Jan Zrimec, Sandra Viknander, JENS KVIST NIELSEN, Aleksej Zelezniak, Martin K. M. Engqvist |
| Citations | 42 (OpenAlex, at import time) |
| References | 86 |
| Model or method | DeepET (deep neural network) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1002/pro.4480 |
| Open access | hybrid |
| Keywords | enzyme, thermal stability, deep learning, protein engineering, DeepET |
Importance rating
significant — thousands to millions affected
Abstract states learning thermal determinants of protein sequences has profound significance for basic biology, drug discovery, and protein engineering; a genuinely new sequence-to-temperature capability, but no scale quantified.