Accurate Solution-Phase Thermodynamics via Composite Quantum Chemistry and Machine Learning: Application to Biomass Decomposition.

Thermodynamic quantities in the hydrated state provide essential reference information for understanding hydrothermal biomass decomposition. However, accurately estimating thermodynamic quantities in the hydrated state using computational methods remains challenging. In this study, we develop a general computational framework to evaluate accurate solution-phase thermodynamics in the standard state for compounds involved in the decomposition of xylose and glucose in water by combining gas-phase free-energy calculations with machine-learning prediction of hydration free energies. Gas-phase thermodynamic quantities are evaluated by a composite method designed to reproduce coupled-cluster theory with singles, doubles, and perturbative triples in the complete basis set limit (CCSD(T)/CBS) accuracy. Hydration free energies are predicted using molecular descriptors and machine learning. The resulting approach enables efficient evaluation of large reaction networks and is demonstrated for a dataset comprising 84 molecules and 73 reactions. Comparison with experimentally-derived free-energy differences for sugar isomerization indicates that the method achieves an accuracy within 3.7 kcal/mol in aqueous solution. The computed thermodynamic data further identify furfural and 5-hydroxymethylfurfural as thermodynamically favorable products in the xylose and glucose systems, respectively. This framework provides a practical route to accurate standard-state solution-phase thermodynamics by combining high-accuracy gas-phase thermochemistry with machine-learning prediction of hydration free energies.