Machine Learning-Driven Multi-Objective Optimization of Enzyme Combinations for Plastic Degradation: An Ensemble Framework Integrating Sequence Features and Network Topology

Plastic waste accumulation presents critical environmental challenges demanding innovative circular economy solutions. This study developed a comprehensive machine learning framework to systematically identify optimal enzyme combinations for polyester depolymerization. We integrated kinetic parameters from the BRENDA database with sequence-derived features and network topology metrics to train ensemble classifiers predicting enzyme-substrate relationships. A multi-objective optimization algorithm evaluated enzyme combinations across four criteria: prediction confidence, substrate coverage, operational compatibility, and functional diversity. The ensemble classifier achieved 86.3% accuracy across six polymer families, significantly outperforming individual models. Network analysis revealed a modular organization with hub enzymes exhibiting broad substrate specificity. Multi-objective optimization identified 156 Pareto-optimal enzyme combinations, with top-ranked pairs achieving composite scores exceeding 0.89. The Cutinase–PETase combination demonstrated exceptional complementarity (score: 0.875±0.008), combining complete substrate coverage with high catalytic efficiency. Validation against experimental benchmarks confirmed enhanced depolymerization rates for recommended enzyme cocktails. This framework provides a systematic approach for enzyme prioritization in plastic valorization, advancing biological recycling technologies through data-driven biocatalyst selection while identifying key economic barriers requiring technological innovation.