Researchers introduce DeltaCata to predict enzyme mutation effects
DeltaCata is a new Siamese geometric graph neural network designed to predict how single‑amino‑acid substitutions change enzyme kinetic parameters such as turnover number (kcat) and Michaelis constant (Km). The authors built a high‑quality mutation‑effect dataset by pairing wild‑type and mutant measurements obtained under identical conditions, using a semi‑automatic pipeline that combines…
Key points
- DeltaCata predicts Δkcat and ΔKm directly from residue substitutions using a Siamese geometric GNN
- Model shows substantial gains over prior methods, especially on unseen enzymes
- Applied to methyltransferase sGmOMT2, achieving up to 7.0‑fold catalytic efficiency improvement
Benchmarks show DeltaCata outperforms existing predictors, especially on enzymes that were not present in the training data. The model’s utility was further demonstrated through in‑silico screening of deep‑mutational‑scanning datasets and by engineering the methyltransferase sGmOMT2, where it enabled up to a 7.0‑fold increase in catalytic efficiency for in‑vitro icaritin synthesis. The study highlights the promise of geometric deep learning for accelerating industrial enzyme engineering.
The work was funded by several Chinese programs, including the Guangdong S&T Program (2024B0101040005) and the National Natural Science Foundation of China (32501105). The article is published under a Creative Commons Attribution‑NonCommercial‑NoDerivatives 4.0 International License.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Researchers link AI agent power to human Goal attainment in formal models · 1 src
- PromptOps framework proposed for iterative LLM judge optimization · 1 src
- Researchers introduce BCSS to scale symbolic models of reasoning · 1 src
- Researchers test Cross-Agent formalization for clinical trial criteria · 1 src
- Researchers introduce efxplorer for Cooperative Multi-Agent reinforcement learning · 1 src
Comments
via GitHub Discussions