DiZyme

DiZyme predicts quantitative catalytic activities of inorganic nanozymes using data-driven machine-learning models and derived descriptors to support nanozyme design and evaluation.

Key Features:

• Extensive Database: A curated database of over 300 inorganic nanozymes compiled from more than 100 scientific articles.
• Data-Driven Predictions: Employs machine learning techniques to predict nanozyme catalytic activities without assuming underlying system parameters.
• Random Forest Regression: Uses a Random Forest Regression model optimized for peroxidase activity with reported R² up to 0.796 for K_cat and 0.627 for K_m.
• Feature Selection and Tuning: Applies correlation-based feature selection and hyperparameter tuning during model development.
• New Descriptors: Introduces descriptors derived from data analysis to identify features responsible for catalytic activity.
• Visualization: Provides visualization tools for interpreting nanozyme data and model results.

Scientific Applications:

• Nanozyme Design: Guides development of customized nanozymes with specific catalytic properties, including peroxidase-like activity.
• Experimental Prioritization: Enables prediction of enzyme activities prior to experimental validation to prioritize experiments and conserve resources.
• Research and Development: Supports academic and industrial research seeking cost-effective and stable alternatives to natural enzymes.

Methodology:

Computational methods comprise literature-based data collection, analysis of existing nanozyme data, derivation of descriptors from data analysis, correlation-based feature selection, and training of Random Forest Regression models with hyperparameter tuning.