Accurate Thermochemistry of Complex Lignin Structures via Density Functional Theory, Group Additivity, and Machine Learning

A molecular-level
understanding of lignin structures and bond dissociation
energies could facilitate depolymerization technologies. Still, this
information is currently limited due to the lack of databases and
the simplification of surrogate models. Here, substitution effects
on seven common linkages in lignin polymers are systematically investigated.
An automated reaction network generator is employed to create a database
of structures. A new group additivity (GA) model based on principal
component analysis (PCA) descriptors is introduced and trained on
gas-phase density functional theory data of 4100 species at the M06-2X/6-311++G­(d,p)
level. Hydrogen bonds, local steric, and nonaromatic ring contributions
are also incorporated. Finally, we improve the accuracy of the group
additivity model to reach the G4 theory by computing a data set of
770 species at this level and using a data fusion approach.