Surface Tension Prediction of Polymers Using Machine Learning and Graph Neural Networks

Surface tension is a critical property influencing polymer behavior at interfaces, impacting applications from coatings to biomedical devices. Traditional experimental methods for measuring surface tension are time-consuming and costly. Molecular dynamics (MD) simulations provide insights but are computationally intensive, especially for long-chain polymers. This study explores machine learning (ML) techniques, including multilinear regression (MLR), random forest (RF), and graph neural networks (GNN), to predict polymer surface tension. A dataset of several hundreds of homopolymers from the PolyInfo database was used to train and evaluate these models. Descriptors were derived at various levels of complexity, from manually calculated features to graph-based representations. While the GNN model outperformed others, simpler models like MLR revealed the impact of specific molecular features on surface tension. Systematic modifications of polymer structures were undertaken, and the predictions were validated against MD simulations, demonstrating that the GNN effectively captured trends in surface tension.