Thermal fluctuation and conformational effects on NMR parameters in β-O-4 lignin dimers from QM/MM and machine-learning approaches. Issue 15 (30th March 2022)

• Record Type: Journal Article
• Authors: Aguilera-Segura, Sonia Milena; Dragún, Dominik; Gaumard, Robin; Di Renzo, Francesco; Ondík, Irina Malkin; Mineva, Tzonka
• Is Part Of: Physical chemistry chemical physics. Volume 24:Issue 15(2022)
• Page Start: 8820
• Page End: 8831
• Publication Date: 2022-03-30
• Subjects: Chemistry, Physical and theoretical -- Periodicals; 541.3
• DOI: 10.1039/d2cp00361a
• Languages: English
• ISSNs: 1463-9076

Abstract

Abstract : Quantum chemical and machine-learning methods may enable high throughput into the structural and dynamics effects on NMR parameters in lignin fragments. Abstract : Advanced solid-state and liquid-state nuclear magnetic resonance (NMR) approaches have enabled high throughput information about functional groups and types of bonding in a variety of lignin fragments from degradation processes and laboratory synthesis. The use of quantum chemical (QM) methods may provide detailed insight into the relationships between NMR parameters and specific lignin conformations and their dynamics, whereas a rapid prediction of NMR properties could be achieved by combining QM with machine-learning (ML) approaches. In this study, we present the effect of conformations of β-O-4 linked lignin guaiacyl dimers on 13 C and 1 H chemical shifts while considering the thermal fluctuations of the guaiacyl dimers in water, ethanol and acetonitrile, as well as their binary 75 wt% aqueous solutions. Molecular dynamics and QM/MM simulations were used to describe the dynamics of guaiacyl dimers. The isotropic shielding of the majority of the carbon nuclei was found to be less sensitive toward a specific conformation than that of the hydrogen nuclei. The largest 1 H downfield shifts of 4–6 ppm were established in the hydroxy groups and the rings in the presence of organic solvent components. The Gradient Boosting Regressor model has been trained on 60% of the chemical environments in the dynamics trajectories with the NMR isotropic shielding ( σ iso ), computed with density-functional theory, for lignin atoms. The high efficiency of this machine-learning model in predicting the remaining 40% σ iso ( 13 C) and σ iso ( 1 H) values was established.