Machine Learning-Based\nHigh-Throughput Screening for\nHigh-Stability Polyimides

High-stability polyimides exhibit tremendous potential\nfor applications\nin flexible electronics, fibers, and membrane materials. However,\nscreening polyimide structures with superior performance remains a\nsignificant challenge. In this study, we combined literature data,\nmachine learning, and molecular dynamics simulations to identify key\nfactors influencing the stability of polyimide structures and screen\nfor high-stability polyimide candidates. Specifically, we utilized\ninterpretable machine learning methods to analyze polyimide systems\ndocumented in the literature, aiming to identify crucial substructures\nthat impact polyimide stability. This approach offers valuable insights\nfor the development of high-stability polymers. By integrating diamine\nand dianhydride structures from both the PubChem database and the\nliterature, we generated a data set containing over 15 million hypothetical\npolyimides. Using appropriate machine learning models, we conducted\nhigh-throughput screening to discover polyimides that simultaneously\nexhibit high thermal stability and excellent mechanical properties.\nThe selected machine learning models demonstrated strong predictive\ncapability in forecasting four key properties: glass transition temperature\n(<i>T</i><sub>g</sub>), Young’s modulus (<i>Y</i><sub>m</sub>), tensile strength (<i>T</i><sub>s</sub>), and elongation at break (<i>E</i><sub>g</sub>). Based on the predictions from the optimal models and synthetic\naccessibility scores, we ultimately identified eight polyimide copolymer\nstructures with outstanding stability, with some of their properties\nvalidated through all-atom molecular dynamics simulations.