Modeling Solvent Effects in Mixed-Solvent Environments for Improved Biomass Conversion
EMSL Project ID
51376
Abstract
The conversion of lignocellulosic biomass is a promising method to produce biofuels and chemicals for further microbial conversion from a renewable carbon resource via liquid-phase, acid-catalyzed reactions. A critical variable that affects reaction rates, product selectivities, and the stability of desired products for these reactions is the solvent composition. For example, reaction rates for the conversion of biomass-derived small molecules can increase by two orders of magnitude in mixtures of water and organic, polar-aprotic cosolvents compared to the same reactions in pure water. However, further molecular-scale insight is needed to understand the physical origin of these solvent effects and guide the selection of solvent compositions to improve reaction performance. We propose to combine atomistic molecular dynamics (MD) simulations and data-centric modeling to investigate the hypothesis that solvent effects on acid-catalyzed reactions can be predicted by characterizing the solvent environment near a reactant without directly modeling reaction mechanisms. This hypothesis transforms the study of liquid-phase reactions -- which are challenging to study with quantum chemical simulation techniques -- to the study of multicomponent solution thermodynamics, which can be assessed with classical MD simulations. Because atomistic MD cannot directly model reactions, we will utilize data-centric regression models to relate MD-derived descriptors to experimentally determined reaction rates and selectivities. This descriptor-based approach will be complemented by using deep learning to efficiently predict solvent effects from MD data without requiring human-guided analysis. Together, these methods will provide physical insight into the mechanisms underlying solvent effects and new computational tools to facilitate the screening of solvent mixtures for efficient biomass conversion. Past and preliminary results indicate that data-centric models can predict both reaction rates and product selectivities in good agreement with experiments, supporting the overall hypothesis of the proposal.
Building upon our prior studies of cellulose model compounds, the proposed modeling approach will be used to achieve three Aims. In Aim 1, we will study solvent effects on the chemical and physical deconstruction of cellulose nanofibrils in solvent mixtures containing both water and polar aprotic cosolvents. In Aim 2, we will study the hydrolysis of lignin model compounds to relate reactant and solvent properties to acid-catalyzed reaction rates. In Aim 3, we will study the depolymerization of lignin oligomers to relate the chemical composition and conformational complexity of lignin to related solvent effects. The size of these simulations requires substantial computational resources, necessitating access to EMSL computing facilities. Outcomes of this project will include: (i) insight into the relationship between reactant properties, solvent properties, solution thermodynamics, and resulting reaction rates and selectivities for biomass-relevant compounds; (ii) insight into the influence of reactant properties on local solvent spatial distributions; (iii) data-centric models to predict acid-catalyzed rates and selectivities for model compounds and biopolymers; (iv) relationships between cellulose crystallinity and deconstruction outcomes; (v) relationships between lignin chemical properties, solvent composition, the conformations of lignin polymers in solution, and depolymerization outcomes; and (vi) new computational tools for screening solvent composition to improve acid-catalyzed reaction performance.
Project Details
Project type
Large-Scale EMSL Research
Start Date
2020-10-01
End Date
2022-09-30
Status
Closed
Released Data Link
Team
Principal Investigator
Team Members