Machine Learning-Based High-Throughput Screening for High-Stability Polyimides

High-stability polyimides exhibit tremendous potential for applications in flexible electronics, fibers, and membrane materials. However, screening polyimide structures with superior performance remains a significant challenge. In this study, we combined literature data, machine learning, and molecular dynamics simulations to identify key factors influencing the stability of polyimide structures and screen for high-stability polyimide candidates. Specifically, we utilized interpretable machine learning methods to analyze polyimide systems documented in the literature, aiming to identify crucial substructures that impact polyimide stability. This approach offers valuable insights for the development of high-stability polymers. By integrating diamine and dianhydride structures from both the PubChem database and the literature, we generated a data set containing over 15 million hypothetical polyimides. Using appropriate machine learning models, we conducted high-throughput screening to discover polyimides that simultaneously exhibit high thermal stability and excellent mechanical properties. The selected machine learning models demonstrated strong predictive capability in forecasting four key properties: glass transition temperature ( T g ), Young’s modulus ( Y m ), tensile strength ( T s ), and elongation at break ( E g ). Based on the predictions from the optimal models and synthetic accessibility scores, we ultimately identified eight polyimide copolymer structures with outstanding stability, with some of their properties validated through all-atom molecular dynamics simulations.