Sequential causal machine learning assisted design of biochar for laccase immobilization

Area of Science:

• Environmental Science
• Biotechnology
• Materials Science

Background:

• Optimizing biochar preparation for enzyme immobilization is challenging due to complex causal relationships.
• Understanding the interplay between preparation conditions, biochar properties, and enzyme activity is crucial.

Purpose of the Study:

• To develop a novel sequential causal machine learning model coupled with reinforcement learning.
• To optimize biochar preparation conditions for enhanced immobilized laccase activity.

Main Methods:

• Utilized Random Forest and Adaptive Boosting for prediction accuracy.
• Employed sequential causal modeling and reinforcement learning for optimization.
• Conducted experimental validation of the optimized biochar.

Main Results:

• Achieved excellent prediction accuracy for biochar properties (R² = 0.72 ± 0.11) and immobilized laccase activity (R² = 0.86 ± 0.12).
• Identified raw material type and specific surface area as key drivers of biochar properties and enzyme activity.
• Optimized biochar design yielded 2.18 U/g immobilized laccase activity using agricultural waste.

Conclusions:

• The developed model effectively analyzes nonlinear causal relationships in biochar preparation.
• This approach enables targeted design of biochar for optimized enzyme immobilization.
• The optimized biochar demonstrates high activity, stability, and effectiveness in removing benzo[a]pyrene.