Under the joint guidance of Researcher Li Liangbin and Researcher Yu Wancheng, master's student Sun Shengyang and others in our team applied machine learning to the field of polymer density prediction and achieved accurate prediction of polymer material density through directed graph neural networks. The related work was published in the Journal of Chemical Physics.
The paper proposes a machine learning framework based on Graph Convolutional Neural Networks (GCNN) for accurate polymer density prediction. The study selected data of 1432 homopolymers from the PoLyInfo database and systematically compared the prediction performance of GCNN with other models such as neural networks, random forests, and XGBoost. The results show that the GCNN model combined with the Directed Message Passing Neural Network (D-MPNN) performs the best, with a mean absolute error (MAE) of 0.0497 g/cm³ and a coefficient of determination (R²) of 0.8097. Experimental verification showed that the predicted density of six polymers was highly consistent with the measured values, with a relative error not exceeding 4.8%. SHAP analysis and t-SNE visualization revealed the relationship between functional groups and density, enhancing the interpretability of the model. This research provides an efficient and accurate computational tool for high-throughput polymer screening, which helps promote the design and discovery of polymer materials.
This work was supported by the National Natural Science Foundation of China (No. U2430213) and the Chinese Academy of Sciences (JZHKYPT-2021-04).
S. Sun, F. Tian, C. Zhao, M. Xie, W. Li, W. Yu, K. Cui, and L. Li, Directed message passing neural networks enhanced graph convolutional learning for accurate polymer density prediction. The Journal of Chemical Physics 163(10),(2025)
Paper Link:https://doi.org/10.1063/5.0281450