A REMD-Integrated Thermodynamic Cycle Approach for Accurate Prediction of Solubility in Binary Solvent Systems

Accurate prediction of solubility in complex mixed-solvent systems remains a major challenge due to the intricate interplay of solute–solvent and solvent–solvent interactions, which give rise to complex interfacial phenomena. In this work, we present a physics-based computational framework that integrates replica exchange molecular dynamics (REMD) with sublimation thermodynamic cycle to enable high-precision solubility prediction. Using 1,1-diamino-2,2-dinitroethylene (FOX-7) as a model compound, we apply this method to six binary solvent systems, each composed of dimethyl sulfoxide (DMSO) and one cosolvent, water, ethanol, acetone, acetonitrile, cyclohexanone, or ethyl acetate. By performing parallel sampling at different temperatures, conformational equilibration in complex solvent environments is significantly accelerated. Additionally, the prediction accuracy of sublimation free energy is enhanced by incorporating crystal phonon mode calculations in place of the empirical 2RT approximation, allowing a more realistic treatment of vibrational contributions. Based on 51 individual measurements across the six binary solvent systems, the approach achieves a mean absolute error of 0.38 log units, reducing prediction error by over 44% compared to conventional thermodynamic models such as SMD and COSMO-RS. Furthermore, this physics-based methodology offers broad applicability and scalability for solubility prediction, solvent screening, and crystallization process design.