CrystalGAT: An AI-Based Intelligent Flexible Crystal Materials Design Computing Platform.

Area of Science:

• Materials Science
• Crystallography
• Computational Chemistry

Background:

• Achieving mechanical flexibility in functional crystals is crucial for flexible smart devices.
• Discoveries of flexible crystals have largely been serendipitous, lacking systematic design principles.
• Machine learning offers a transformative approach to materials science research.

Purpose of the Study:

• To develop an innovative platform, CrystalGAT, for predicting the mechanical properties of molecular crystals.
• To enable the rational design of flexible molecular crystals with desired properties.
• To identify key structural features influencing crystal mechanical behavior.

Main Methods:

• Development of CrystalGAT, a graph neural network model utilizing attention mechanisms.
• Implementation of data augmentation strategies for robust model construction.
• Validation of prediction accuracy and generalization capabilities on diverse crystal systems.

Main Results:

• CrystalGAT achieved a 90% prediction accuracy on the validation set.
• The model demonstrated generalization to multicomponent crystal systems.
• Key crystal segments influencing mechanical properties were identified, enabling the transformation of brittle to flexible crystals.
• Efficient screening of plastic drug crystals for enhanced tableting performance was achieved.

Conclusions:

• CrystalGAT provides an efficient and accurate method for flexible molecular crystal design.
• The platform has potential applications in material discovery and drug molecular modification.
• This work facilitates the development of novel flexible functional materials and optimized drug formulations.