(623b) Development and Validation of a New Virtual Coformer Screening Model

A novel virtual coformer screening model, COSMO-RS+Δ-ML, integrating COSMO-RS with machine learning to account for both miscibility in the amorphous phase and crystallinity contributions to cocrystallization, was proposed. This computational approach was validated against published experimental cocrystal screening data for multiple APIs, demonstrating superior performance compared to the pure COSMO-RS method. The COSMO-RS+Δ-ML model was subsequently applied to guide targeted experimental cocrystal screenings for tiopronin and dapagliflozin. This led to the discovery and characterization of multiple cocrystal hits, including the first known anhydrous cocrystal of tiopronin.