Deep Learning Model for Prediction of Entanglement Molecular Weight of Polymers

Abstract

Entanglement molecular weight is one of the key polymer properties strongly related to many mechanical and dynamic behaviors of polymers. Despite its importance, the data for entanglement molecular weight by either measurements or predictions are still far from covering a wide range of polymer species. To address this issue, we employed the deep learning technique to predict the entanglement molecular weight of polymers using graph convolutional neural networks that convert molecules into graph structures. In addition, to overcome the limitation due to the lack of data, the transfer learning technique, which transfers knowledge learned through large-scale datasets, was also introduced to improve the performance. The trained neural network model showed higher prediction performance than the conventional prediction methods.