Accelerated discovery of highly active enzyme nanohybrids with parallelized Bayesian optimization in hybrid space

Area of Science:

• Biotechnology
• Protein Engineering
• Machine Learning

Background:

• Enzyme immobilization is crucial for industrial applications but often reduces activity due to diffusion limitations.
• Optimizing immobilized carriers typically requires extensive trial-and-error experiments.
• Fragile biomacromolecules need protective carriers for stability under harsh conditions.

Purpose of the Study:

• To develop a machine-learning-guided workflow to accelerate the discovery of enzyme nanohybrids.
• To optimize nanocarriers for specific enzymes and reactions, enhancing catalytic activity.
• To overcome the limitations of traditional trial-and-error methods in enzyme immobilization.

Main Methods:

• Implementation of a machine-learning workflow utilizing parallelized hybrid-space Bayesian optimization (PHBO).
• Leveraging prior knowledge, machine learning, and iterative feedback for data-efficient optimization.
• Exploration of a vast reaction space (over 10^7 experiments) within a limited experimental budget.

Main Results:

• Achieved 100% activity recovery for glucose oxidase.
• Demonstrated 90% activity recovery for catalase.
• Obtained 79% activity recovery for Candida Antarctica lipase B.
• Successfully identified high-activity enzyme nanohybrids across diverse enzymatic systems.

Conclusions:

• Data-efficient optimization significantly accelerates the discovery of enzyme nanohybrids.
• The PHBO algorithm and ML-guided workflow are effective for diverse enzymatic systems.
• This approach enhances enzyme stability and activity for industrial biocatalysis.