High Glass Transition Temperature Fluorinated Polymers Based on Transfer Learning with Small Experimental Data.

Area of Science:

• Polymer Science
• Materials Science
• Computational Chemistry

Background:

• Machine learning (ML) models can predict polymer properties and explore chemical spaces.
• Limited experimental data restricts the predictive accuracy of ML models for polymers.
• Developing high glass transition temperature (T_g) polymers is crucial for advanced applications.

Purpose of the Study:

• To develop an efficient ML approach for predicting the T_g of fluorinated polymers.
• To guide the design of high T_g copolymers using predictive modeling.
• To overcome challenges associated with small and unevenly distributed polymer datasets.

Main Methods:

• Leveraging transfer learning by pretraining on the QM9 dataset for robust molecular representations.
• Employing ensemble modeling to enhance prediction robustness and reliability.
• Fine-tuning the model on a specialized, limited copolymer dataset.

Main Results:

• The ML model successfully predicted 247 potential high T_g fluorinated polymer candidates (T_g > 390 K).
• 14 of the predicted high T_g candidates were experimentally validated.
• The study demonstrated effective navigation of a large chemical space (61 monomers).

Conclusions:

• Transfer learning and ensemble modeling are effective strategies for ML in polymer design with limited data.
• The proposed approach accelerates the discovery of novel high T_g fluorinated polymers.
• ML holds significant potential for advancing materials design and discovery.