Machine learning for mechanistic insights and optimization in CO₂ cycloaddition catalysis
Document Type
Review
Publication Date
1-25-2026
Abstract
Converting CO₂ into cyclic carbonates via cycloaddition with epoxides is a key catalytic process for sustainable chemical synthesis and carbon mitigation, with 100 % atom economy. Machine learning (ML) drives catalyst design, reaction optimization, and mechanistic insights, achieving predictive accuracies up to R² = 0.99. This review (2020–2025) covers ionic liquids, metal-organic frameworks, and single-atom catalysts, achieving > 90 % yields at ambient conditions with activation energies of 10–20 kcal/mol. Despite challenges like dataset biases, the novel UniDesc-CO2 framework scales datasets to > 10,000 entries using standardized descriptors and active learning. Explainable AI (e.g., SHAP) clarifies descriptors like anion nucleophilicity, advancing sustainable CO₂ cycloaddition catalysis for scalable processes.
Publication Title
Applied Catalysis A General
ISSN
0926860X
DOI
10.1016/j.apcata.2025.120679
Recommended Citation
Abdul Wahab, Yasmin; Shapril, Nur’ain Nadia; Johari, Suzaimi; and Johan, Mohd Rafie, "Machine learning for mechanistic insights and optimization in CO₂ cycloaddition catalysis" (2026). Research Publications (2026 to 2030). 233.
https://knova.um.edu.my/research_publications_2026_2030/233
Volume
710