POS10-1309
Data-Driven Screening of Anion Exchange Membranes Using Quantum Chemical and Structural Descriptors of Extended Ion Exchange Groups
Topic
S10. AI-assisted Design and Simulation of Polymers
When and Where
Sep 30, 2026
08:30 - 09:30
Room 301 (Grand Ballroom)
Session Chairs
Heesuk KIM
Jinhye BAE
Presenter(s)
Nana Terasoba (Tokyo Metropolitan University)
Co-Author(s)
Abstract
Anion exchange membrane (AEM) water electrolysis for green hydrogen production requires membranes that combine high OH⁻ conductivity with alkaline durability. However, AEM structure-property relationships involve many interacting factors, making it difficult to establish a systematic understanding of overall membrane performance. Machine learning methods that encode the full polymer structure as string-based descriptors require increasingly large datasets as structural diversity grows, and the polymers' electronic properties are not directly represented by these descriptors. Here we focus on the extended ion exchange group (eIEG), the local structural unit consisting of the ion exchange group, the neighboring substituents, and the spacer. Fifty AEMs across twenty-six studies were compiled into a property database covering ion conductivity and alkaline stability, defined as the time for conductivity to drop to 90% of its initial value. For each membrane, the eIEG was isolated and characterized using RDKit-derived structural descriptors, including weight fraction and ring count, together with DFT-derived quantum chemical descriptors, such as the HOMO-LUMO gap and dipole moment. Random Forest models, built on these descriptors and interpreted using SHAP analysis showed that both descriptor families contribute to ion conductivity and alkaline stability. Quantum chemical descriptors play an important role in alkaline stability, suggesting a connection between eIEG electronic structure and alkaline durability. Because these trends emerge at the fragment level without modeling the full polymer chain, eIEG-based descriptors provide a practical basis for screening candidate structures before polymer synthesis. We are currently extending the descriptor set and evaluating its applicability across a broader range of polymer chemistries, ultimately aiming to establish more general design guidelines while reducing experimental cost.
