Glass Transition Temperature Prediction of Polymers via Graph Reinforcement Learning.

Area of Science:

• Polymer Science
• Materials Informatics
• Computational Chemistry

Background:

• Graph-based models are widely used for predicting polymer structure-property relationships.
• Existing methods often underutilize unsupervised structural information, limiting predictive accuracy.

Purpose of the Study:

• To develop an advanced model for predicting the glass transition temperature (Tg) of heterocyclic polymers, focusing on polyimides.
• To fully exploit both global and local molecular structural information for improved Tg prediction.

Main Methods:

• Proposed a novel graph reinforcement learning framework: Molecular Structural Regularized Graph Convolutional Network with Reinforcement Learning (MSRGCN-RL).
• Integrated Graph Neural Networks (GNNs) with reinforcement learning (RL) for polymer property prediction.

Main Results:

• Demonstrated the critical importance of both global and local structural regularization for precise Tg prediction.
• Showcased the effectiveness of optimizing MSRGCN training through RL for enhanced performance.

Conclusions:

• The MSRGCN-RL framework significantly improves the prediction accuracy of polymer glass transition temperatures.
• This research pioneers the integration of GNNs and RL for advanced polymer property prediction.