PolyT-GNN: A Graph Neural Network Framework for Data-Driven Discovery of High-Temperature Two-Way Shape Memory...

Area of Science:

• Materials Science
• Polymer Science
• Computational Chemistry

Background:

• Two-way shape memory polymers (2W-SMPs) offer reversible actuation for advanced applications but their discovery is hindered by limited data and predictive models.
• High-temperature 2W-SMPs are particularly challenging to design due to complex structure-property relationships.

Purpose of the Study:

• To introduce PolyT-GNN, a graph neural network framework for the data-driven discovery of high-temperature 2W-SMPs.
• To enable accurate prediction of polymer transition temperatures and facilitate the design of novel 2W-SMP formulations.

Main Methods:

• Compiled a dataset of 170 experimentally validated 2W-SMPs from existing literature.
• Developed PolyT-GNN integrating atomic, bond, molecular descriptors, and monomer weight ratios for transition temperature prediction.
• Employed pretraining and fine-tuning strategies to enhance model accuracy and transferability.

Main Results:

• Achieved a test R-squared of 0.84 in predicting transition temperatures, even with limited data.
• Demonstrated a 38% improvement in prediction accuracy using pretraining and fine-tuning.
• Generated and screened over 80,693 new 2W-SMP formulations.
• Synthesized and validated a polyethylene-dicumyl-peroxide system with a melting transition near 130 °C, closely matching the predicted 113.25 °C.

Conclusions:

• PolyT-GNN provides a robust framework for linking molecular structure, composition, and actuation behavior in 2W-SMPs.
• The data-driven approach enables rational design of high-temperature 2W-SMPs for next-generation smart materials.
• This work significantly advances the discovery and application of advanced shape memory polymers.