AI-Powered Knowledge Base Enables Transparent Prediction of Nanozyme Multiple Catalytic Activity

Area of Science:

• Materials Science
• Computational Chemistry
• Biotechnology

Background:

• Nanozymes possess diverse properties valuable for biomedicine, biosensing, and environmental monitoring.
• Discovering novel nanozymes with predictable catalytic activities is crucial for expanding their applications.

Purpose of the Study:

• To develop a machine learning (ML) approach for identifying new nanozymes.
• To create a comprehensive web platform (DiZyme) for nanozyme research, including a database and predictive tools.
• To enable accurate prediction of multiple nanozyme catalytic activities.

Main Methods:

• Developed an ensemble learning algorithm for predicting nanozyme catalytic activities.
• Compiled a database of 1210 experimental nanozyme samples.
• Integrated a large language model (LLM) powered assistant (DiZyme Assistant) for accessing experimental details.

Main Results:

• Achieved high prediction accuracy for Michaelis-Menten constant (R² = 0.75) and maximum velocity (R² = 0.77) on unseen data.
• The DiZyme platform provides a state-of-the-art database and predictive capabilities for nanozymes.
• The DiZyme Assistant offers supporting information on synthesis and measurement protocols.

Conclusions:

• The developed ML approach and DiZyme platform facilitate the discovery and application of novel nanozymes.
• Accurate prediction of multiple catalytic activities (peroxidase, oxidase, catalase) is now feasible.
• DiZyme promotes advancements in surface-modified inorganic nanozymes for various scientific fields.