Machine Learning Screening of Efficient Ionic Liquids for Targeted Cleavage of the β–O–4 Bond of Lignin

Lignin conversion into high value-added chemicals is of great significance for maximizing the use of renewable energy. Ionic liquids (ILs) have been widely used for targeted cleavage of the C–O bonds of lignin due to their high catalytic activity. Studying the cleavage activity of each IL is impossible and time-consuming, given the huge number of cations and anions. Currently, the mainstream approach to determining the cleavage activity of one IL is to calculate the activation barrier energy (Ea) theoretically via transition state search, a process that involves the iterative determination of an appropriate "imaginary frequency". Machine learning (ML) has been widely used for catalyst design and screening, enabling accurate mapping from specified descriptors to target properties. To avoid complicated Ea calculations and to screen potential candidates, in this study, we selected nearly 103 ILs and guaiacylglycerol-β-guaiacyl ether (GG) as the lignin model and used the ML technology to train models that can rapidly predict the cleavage activity of ILs. Taking the easily accessible bond dissociation energy (BDE) of the β–O–4 bond in GG as the target, an ML model with r > 0.93 for predicting the catalytic activity of ILs was obtained. The change tendency of the BDE is consistent with the experimental yield of guaiacol, reflecting the reliability of the ML model. Finally, [C2MIM][Tyrosine] and [C3MIM][Tyrosine] as the optimal candidates for future applications were screened out. This is a novel strategy for predicting the catalytic activity of ILs on lignin without the need to calculate complicated reaction pathways while reducing time consumption. It is anticipated that the ML model can be utilized in future practical applications for targeted cleavage of lignin.