A New Hybrid Virtual Screening Model for Predicting Solvate Formation Propensity in Pharmaceutical Crystallization

Efficient solvent selection is essential for controlling crystalline solvate formation in pharmaceutical crystallization. A novel COSMO-RS+ML hybrid virtual solvent screening model has been developed to predict solvate propensity by integrating COSMO-RS amorphous-phase miscibility calculations with machine learning predictions of crystallinity contributions. Validated against ten reported studies involving prolific solvate-forming molecules, the model significantly outperformed the pure COSMO-RS method in ranking solvents by their likelihood of forming solvates. When combined with targeted experimental follow-up, this computational approach provides a powerful strategy to accelerate and derisk solvent selection in pharmaceutical crystallization development.