Machine learning prediction of photocatalytic lignin cleavage of C–C bonds based on density functional theory

Photocatalytic degradation is a promising method for producing high-value chemicals from lignin through the cleavage of targeted chemical bonds. In this study, machine learning combined with density functional theory (DFT) was used to analyze lignin structure and offer insight and guidance for the design of active and selective photocatalytic C-C cleavage systems for lignin valorization under mild conditions. Classification training revealed that the random forest (RF) model provided the highest test accuracy (accuracy score: 0.99) compared with those of the K-nearest neighbor (K-NN), naïve Bayes (NB), support vector machine (SVM), and logistic regression (LR) models. The dissociation energy for bond breakage was found to increase as the number of methoxy groups attached to the benzene rings increased. The reaction conditions were found to contribute 39.22% to model feature importance, and that oxygen is an important atmospheric component for the photocatalytic degradation of lignin. In addition, the specific surface area of the catalyst can be used as an important screening index. • Lignin descriptors were introduced with the aid of density functional theory. • Random forest model predicted C-C/C-O bond breakage with an accuracy of 0.99. • Random Forest was used to determine the feature importance of broken C-C bonds. • Partial dependence analysis diagram provides a reference for breaking C-C bonds.