Fig. 1: Schematics of the process integrating high-throughput MD and machine learning to explore high-viscosity–temperature performance polymers.

a Overview of the workflow, including data collection and generation, feature engineering, virtual screening, and explainable AI; b Detailed technical pathway: The VIIInfo dataset was generated through automated high-throughput MD. Descriptor selection was performed using correlation coefficients and Recursive Feature Elimination (RFE). High-throughput virtual screening under multiple constraints was conducted using accurate algorithms (XGBoost, KRR, MLP, RF). Finally, quantitative structure–property relationships (QSPR) were developed and analyzed using interpretable SHAP and symbolic regression (SR).