Machine Learning Accelerates Crystallization for Structure Determination

Single-crystal X-ray diffraction (SCXRD) is a powerful tool for structural elucidation, but requires high-quality crystals that are often difficult to obtain. The crystalline mate strategy helps overcome this limitation by facilitating the co-crystallization of volatile or complex molecules, with few restrictions on size or purity. However, defining its scope of applicability remains challenging, until now requiring experimental trial-and-error screening. Here, we demonstrate a machine learning (ML)-accelerated workflow that rapidly identifies suitable candidates for co-crystallization. Through feature engineering and workflow optimization, we trained the MCC model, achieving over 95% prediction accuracy. Experimental validation confirmed 114 successful co-crystals among 120 predicted compounds. The wide structural and functional diversity exhibited highlights the robustness and broad applicability of our strategy, enabling efficient discovery of new structures by SCXRD under standard laboratory conditions.