Physics-informed, dual-objective optimization of high-entropy-alloy nanozymes by a robotic AI chemist

Abstract

Engineering artificial nanozymes as substitutes for natural enzymes presents a significant challenge. High-entropy alloys (HEAs) show great promise for mimicking peroxidase (POD) activity, yet discovering HEAs that surpass the catalytic efficiency of natural horseradish peroxidase (HRP) remains a formidable task. In this study, we developed a robotic artificial intelligence chemist integrating theoretical calculations, machine learning, Bayesian optimization (BO), and on-the-fly data analysis by a large language model (LLM). At the core of our approach is a physics-informed, multi-objective optimization framework that simultaneously optimizes multiple key properties of nanozymes. By incorporating an auxiliary knowledge model and leveraging collaborative LLM-in-the-loop feedback, we significantly enhanced the BO process, accelerating the data-driven discovery. This integrated approach outperformed both random sampling and standard BO, enabling efficient exploration of the vast chemical space and the identification of HEAs with POD-mimicking properties that exceed those of the natural enzyme and previously reported HEA and single-atom catalysts.