The catalytic conversion of lignocellulosic biomass is a promising method to produce liquid fuels and chemicals from a renewable carbon resource. Biomass conversion is facilitated by liquid-phase, acid-catalyzed dehydration and hydrolysis reactions that are hindered in aqueous solution by low reactivity and poor selectivity. However, modifying the solvent composition by mixing water with polar aprotic cosolvents can improve the rates of these reactions by 100-fold compared to rates in water. In this talk, I will discuss our efforts to combine molecular simulations with machine learning to understand and predict the influence of solvent composition on the acid-catalyzed conversion of biomass-derived reactants. Classical molecular dynamics simulations reveal that polar aprotic cosolvents drive the formation of local solvent domains that effectively confine hydrophilic reactants and solvating water molecules, enhancing reactivity by increasing reactant-proton association and stabilizing charged transition states. By characterizing properties of the local solvent domain, we can quantitatively predict experimental reaction rates without modeling the reaction mechanism. We further find that the spatial configurations of solvent molecules confined within the local solvent domain influence reaction selectivity by stabilizing specific reactive sites. Based on these insights, we developed a machine learning model to rapidly predict reaction rates by analyzing the complex spatial features encoded within molecular simulation data. The model accurately generalizes to reactants/solvents beyond those in the initial training data, enabling reaction rate predictions for a design space that is too large to study with conventional quantum mechanical methods. These computational models thus provide insight into how confined environments can emerge from the interplay of reactant and solvent interactions in solvent mixtures and enable the efficient screening of solvent systems for improved biomass conversion processes.

Molecular and data-centric modeling of acid-catalyzed reactions in mixed-solvent environments

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.