From Drug Molecules to Thermoset Shape Memory Polymers: A Machine Learning Approach.

Area of Science:

• Materials Science
• Polymer Chemistry
• Machine Learning Applications

Background:

• Ultraviolet (UV)-curable thermoset shape memory polymers (TSMPs) are crucial for 3D/4D printing lightweight, load-bearing structures.
• A key challenge is achieving high recovery stress without excessively high glass transition temperatures (Tg), which can approach decomposition temperatures.
• Existing machine learning (ML) approaches face limitations due to scarce data specific to TSMPs.

Purpose of the Study:

• To develop an enhanced ML approach for discovering novel UV-curable TSMPs with desirable properties, specifically high recovery stress and mild Tg.
• To overcome the data scarcity challenge in TSMP discovery by leveraging information from data-rich sources.
• To create a robust ML framework capable of predicting TSMP properties and exploring a vast compositional space.

Main Methods:

• An enhanced ML approach combining transfer learning-variational autoencoder with a weighted-vector combination method was employed.
• Drug molecule datasets were used for pretraining to map TSMP features into a more Gaussian-like hidden space.
• The framework was designed to effectively handle molar ratio information and learn from the relatedness between data-scarce TSMP data and data-abundant drug data.

Main Results:

• Five new types of UV-curable TSMPs with desired properties were successfully discovered.
• One novel TSMP was experimentally validated, confirming the efficacy of the ML approach.
• The developed ML framework demonstrated superior accuracy and robustness compared to traditional methods like support vector machines.

Conclusions:

• The study presents a state-of-the-art ML framework for discovering novel thermoset shape memory polymers, overcoming significant data limitations.
• The transfer learning approach effectively utilizes related data sources to improve model performance for data-scarce materials.
• This framework opens new avenues for the discovery of advanced thermoset polymers beyond TSMPs for various applications.