Machine Learning Accelerates Crystallization for Structure Determination.

Area of Science:

• Crystallography
• Materials Science
• Computational Chemistry

Background:

• Single-crystal x-ray diffraction (SCXRD) is crucial for structural elucidation but hindered by the difficulty of obtaining high-quality crystals.
• The crystalline template strategy aids co-crystallization of challenging molecules, yet its applicability scope requires experimental screening.

Purpose of the Study:

• To develop a machine learning (ML)-accelerated workflow for predicting suitable co-crystallization candidates.
• To overcome the limitations of trial-and-error screening in identifying co-crystal formation.

Main Methods:

• Feature engineering and workflow optimization were employed to train a machine learning model (MCC model).
• The model was trained on data to predict the success rate of co-crystallization experiments.

Main Results:

• The MCC model achieved over 95% prediction accuracy for co-crystallization candidates.
• Experimental validation confirmed 114 successful co-crystals out of 120 predicted compounds, demonstrating high reliability.
• The strategy showed broad applicability across diverse structures and functions.

Conclusions:

• The ML-accelerated workflow enables rapid and efficient identification of co-crystallization candidates.
• This approach significantly enhances the discovery of new structures via SCXRD under standard laboratory conditions.