Enhanced Sequence-Activity Mapping and Evolution of Artificial Metalloenzymes by Active Learning

Area of Science:

• Biocatalysis
• Protein Engineering
• Machine Learning in Chemistry

Background:

• Enzyme engineering is vital for sustainable bioeconomy but is often inefficient and relies on chance.
• Current methods struggle to model large protein sequence spaces effectively within practical experimental constraints.

Purpose of the Study:

• To develop and validate an integrated pipeline combining large-scale screening and active machine learning to enhance enzyme engineering efficiency.
• To engineer a novel artificial metalloenzyme (ArM) for a new hydroamination reaction using the developed pipeline.

Main Methods:

• Implemented a pipeline integrating lab automation, next-generation sequencing, and active machine learning (Gaussian process regression).
• Generated sequence-activity data for thousands of ArM variants, incorporating an explorative screening round and accounting for experimental noise.
• Utilized Gaussian process regression to model the sequence-activity landscape and guide iterative screening.

Main Results:

• Achieved an order-of-magnitude increase in the hit rate for identifying improved enzyme variants.
• Demonstrated efficient use of experimental resources through data-informed screening.
• Successfully engineered an artificial metalloenzyme for a challenging hydroamination reaction.

Conclusions:

• The integrated active machine learning and large-scale screening pipeline significantly improves the efficiency and success rate of enzyme engineering.
• This approach accelerates the development of novel biocatalysts, offering broad utility for sustainable chemistry and bioeconomy applications.