Explicit microsolvation as a "conformational chaperone" for openCOSMO-RS solvation energy predictions.

Abstract

Accurate prediction of solvation free energies is essential for modeling phase partitioning and chemical behavior, a process fundamentally dependent on identifying relevant solution-phase molecular conformations. Standard implicit solvation models, while efficient, often fail to capture specific solute-solvent interactions that dictate conformational preference, particularly when strong intramolecular hydrogen bonds are present. In this work, we demonstrate that incorporating explicit solvation during conformational sampling reduces systematic errors in subsequent openCOSMO-RS predictions. Using the default openCOSMO-RS workflow as well as the global optimization algorithm (GOAT) and commandline energetic sorting (CENSO) workflow, we show that the inclusion of a single explicit water molecule acts as a "conformational chaperone", disrupting stable gas-phase intramolecular interactions and accessing geometries that are more consistent with solvent-exposed solution-phase conformations. Analysis of the resulting COSMO-RS σ-profiles reveals that these microsolvated geometries consistently expose critical polar surface area, translating even subtle structural shifts into significant thermodynamic corrections. For complex hydrogen-bond donor solutes in the SM8 dataset, this method yields an accuracy improvement, reducing the absolute root-mean-square error (RMSE) by 7.3% to 13.0% depending on the workflow. Furthermore, by identifying consistent structural motifs induced by the chaperone, such as systematic N-H and O-H bond elongations, we demonstrate that artificially stretching these bonds in non-solvated conformers successfully mimics some of the benefits of explicit solvation at negligible computational cost. While higher-order microsolvation (n > 1) yields diminishing returns due to solvent self-interaction, single-molecule microsolvation offers a physically rigorous, cost-effective strategy for correcting structural bias in environmental property predictions.