DeepEnzyme: a robust deep learning model for improved enzyme turnover number prediction by utilizing features of protein 3D-structures
Problem
Experimentally measuring enzyme turnover numbers (kcat) is time-consuming, and existing deep learning predictors lack accuracy and robustness especially for enzymes with low sequence similarity to training data.
Why it matters
kcat values are essential for protein engineering and synthetic biology, and slow experimental measurement limits research progress.
What the AI did
DeepEnzyme combines Transformer and Graph Convolutional Network architectures using both protein sequence and 3D-structure features to predict kcat.
Results
not stated in source
Impact
The abstract claims improved robustness for low-similarity enzymes but provides no concrete metrics.
Record
| Venue | Briefings in Bioinformatics |
|---|---|
| Published | 2024 |
| Work type | article |
| Authors | Tong Wang, Guangming Xiang, Siwei He, Liyun Su, Yuguang Wang, Xuefeng Yan, Hongzhong Lu |
| Citations | 56 (OpenAlex, at import time) |
| References | 53 |
| Model or method | DeepEnzyme (Transformer + GCN) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1093/bib/bbae409 |
| Open access | gold |
| Keywords | enzyme turnover number, kcat prediction, deep learning, Transformer, GCN, protein 3D-structure |
Importance rating
real but narrower advance
Experimental kcat measurement is time-consuming and blocks protein engineering and synthetic biology; abstract claims wide applications but gives no scale or metrics; enabling but narrower advance.