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

Abstract Artificial intelligence (AI) has significantly advanced protein engineering, enabling rapid enzyme evolution for diverse applications. However, the fragile nature of biomacromolecules requires enzyme immobilization to preserve catalytic activity under harsh industrial conditions, which often restricts substrate diffusion and reduces enzymatic activity. This challenge demands extensive trial-and-error experiments to optimize immobilized carriers with high activity for different enzymes. Here we show a machine-learning-guided workflow along with an algorithm named parallelized hybrid-space Bayesian optimization (PHBO) to accelerate the discovery of nanocarriers for specific enzymes and reactions. Leveraging prior knowledge, machine learning and iterative feedback, within limited number of experiments, this workflow explores the reaction space of over 10 7 experiments and achieves activity recovery of 100%, 90%, and 79% for glucose oxidase, catalase, and Candida Antarctica lipase B, respectively. These results demonstrate that data-efficient optimization can substantially accelerate the discovery of enzyme nanohybrids with high catalytic activity across diverse enzymatic systems.