Machine Learning Guided Graphdiyne based Nanozyme Discovery

Abstract

Nanozymes are nanomaterials with enzyme-like activities, which are promising alternatives for natural enzymes because of their features of excellent performance, controllable activity, and environmental resistance. Improving the catalytic activity and broadening the scope of nanozymes are prerequisites to supplementing or even superseding enzymes. However, the discovery of nanozymes is mostly relied on serendipity with limited fine-tunings of chemical composition, which is often incomprehensive and fragmented so that severely impedes the development of novel nanozymes. In this work, a comprehensive investigation of the peroxidase mimic reaction catalyzed by nonmetallic doped-graphdiyne (GDY) is presented in terms of the dopant atom electronegativity, atomic radius, elemental doping sites, hydrogen peroxide adsorption sites, and bandgaps, covering all nonmetallic elements and enabling the screening of potential nanozymes. The density functional theory (DFT) calculations carefully explore the properties of nanozymes by analyzing two key steps of Gibbs’s free energy of H2O2 decomposition reaction. The machine learning (ML) technique AdaBoost integration model is innovatively utilized based on a fuzzy model for data predicting and it can predict the nanozyme performance with high regression accuracy (R2 >93%) in test set. Thereafter, six different predicted performances were synthesized are synthesized and confirmed in accordance with their excellent kinetics parameters experimentally. From three independent perspectives (DFT, ML, and experimental validation), this work proposes pivotal guidelines for nanozyme design and shows great potential as a general approach for future nanozyme discovery.