Integrated design of solvent–antisolvent mixtures and crystallization processes powered by machine learning

Crystallization is a key separation technology in the chemical and pharmaceutical industries, offering high-purity products with relatively low energy consumption. However, the design of efficient antisolvent crystallization processes is inherently complex due to the interactions between solvent and antisolvent, as well as the selection of process conditions. Existing computer-aided molecule and process design (CAMPD) frameworks rely on group contribution or quantum-mechanical methods for thermophysical property predictions, which either limit the molecular design space or result in high computational costs. To overcome these challenges, we couple machine learning-based property predictions with a SMILES-based molecular design algorithm into a CAMPD framework (ML-CAMPD), enabling rapid and accurate solvent selection for crystallization. We demonstrate this ML-CAMPD framework through the case study of ibuprofen antisolvent crystallization, showing improvements in process efficiency. A screening study identified acetone-water as the most promising solvent–antisolvent pair. By applying the CAMPD framework to design new mixtures, we find solvent–antisolvent systems that outperformed acetone-water by 10% in energy efficiency. The proposed approach broadens the applicability of CAMPD frameworks and offers a powerful tool for designing efficient and sustainable crystallization processes.