Presentation Information
[2ASBA-06]Data-Driven Enzyme Design for Rapid Development of Sustainable Oxidative Biocatalysis
○Ee Lui Ang1 (1. Singapore Institute of Food and Biotechnology Innovation (Singapore))
Keywords:
Oxidative enzyme engineering,Multi-modal modelling,Machine-learning accelerated biocatalyst development
The pace of enzyme engineering remains a key bottleneck in deploying biocatalysis for sustainable chemical manufacturing, particularly for reactions involving complex, industrially relevant substrates. Here, we present a data-driven framework that accelerates oxidative biocatalyst development by integrating directed evolution, multi-modal modelling, and machine learning into a design–predict–build workflow.
Using galactose oxidase (GOase) as a model system, we demonstrate unprecedented expansion of substrate scope toward bulky benzylic and secondary alcohols, achieving up to 2,400-fold activity improvements, alongside enhanced thermostability and relaxed substrate enantioselectivity. We established predictive models that quantitatively link enzyme sequence, structure, and function. These models enable accurate enzyme–substrate matching without additional training and reduced experimental screening effort. We further validate real-world applicability through rapid identification of optimal GOase variants for the synthesis of the pharmaceutical intermediate Dyclonine.
To demonstrate generality, we extend this framework to unspecific peroxygenases (UPOs), where a combined pipeline of in silico sequence mining, heterologous expression, and machine learning yields predictive accuracies up to 76% across diverse substrates, significantly lowering barriers to UPO deployment. Collectively, this work establishes a generalizable, predictive paradigm for enzyme engineering that transforms the exploration of enzyme sequence–function space from a screening-limited process into a design-driven capability. By drastically reducing experimental burden while unlocking new chemical space, this approach enables faster translation of biocatalysis into industrial applications and highlights the transformative potential of AI-enabled synthetic biology.
Using galactose oxidase (GOase) as a model system, we demonstrate unprecedented expansion of substrate scope toward bulky benzylic and secondary alcohols, achieving up to 2,400-fold activity improvements, alongside enhanced thermostability and relaxed substrate enantioselectivity. We established predictive models that quantitatively link enzyme sequence, structure, and function. These models enable accurate enzyme–substrate matching without additional training and reduced experimental screening effort. We further validate real-world applicability through rapid identification of optimal GOase variants for the synthesis of the pharmaceutical intermediate Dyclonine.
To demonstrate generality, we extend this framework to unspecific peroxygenases (UPOs), where a combined pipeline of in silico sequence mining, heterologous expression, and machine learning yields predictive accuracies up to 76% across diverse substrates, significantly lowering barriers to UPO deployment. Collectively, this work establishes a generalizable, predictive paradigm for enzyme engineering that transforms the exploration of enzyme sequence–function space from a screening-limited process into a design-driven capability. By drastically reducing experimental burden while unlocking new chemical space, this approach enables faster translation of biocatalysis into industrial applications and highlights the transformative potential of AI-enabled synthetic biology.
Comment
To browse or post comments, you must log in.Log in