Machine Learning Prediction of Laccase-Catalyzed Oxidation of Aromatic Compounds Using Curated Enzyme-Specific...

Area of Science:

• Biocatalysis
• Green Chemistry
• Enzyme Engineering

Background:

• Laccases are multi-copper oxidase enzymes utilized as biocatalysts in green chemistry due to their ability to oxidize diverse substrates with water as the only byproduct.
• Predicting laccase-substrate compatibility is challenging due to complex interactions involving enzyme structure, properties, redox potential, and environmental factors.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting laccase-substrate interactions.
• To streamline experimental workflows for identifying suitable laccase-substrate combinations for biocatalysis.

Main Methods:

• Application of four classical machine learning classifiers and a transformer-based model (ChemBERTa).
• Evaluation on three curated datasets of aromatic substrates and their oxidation profiles with specific laccases.
• Analysis of feature importance (Random Forest) and attention mechanisms (ChemBERTa) to identify key molecular determinants of oxidation.

Main Results:

• Machine learning models demonstrated comparable performance in predicting laccase-substrate compatibility.
• Random Forest (RFC) exhibited superior stability across different data splits and laccases.
• Feature importance and attention analyses identified critical molecular features influencing oxidation outcomes.
• A tool for visualizing ChemBERTa predictions by mapping SMILES attributions onto molecular graphs was developed.

Conclusions:

• Machine learning provides a robust and interpretable framework for accelerating the discovery of laccase-substrate pairs.
• RFC models offer stable and reliable predictions for laccase-substrate screening.
• Understanding molecular features aids in rational enzyme engineering and substrate design for enhanced biocatalysis.