Uncovering Structure–Conductivity Relationships in AEMs Using Interpretable Machine Learning

Authors:
Preprint
Publisher website
Google Scholar
Abstract
Anion exchange membranes (AEMs) play a vital role in the performance of water electrolyzers and fuel cells, yet their discovery and optimization remain challenging due to the complexity of structure–property relationships. In this study, we introduce a machine learning framework that leverages both graph neural networks (GNNs) and descriptor-based models to predict and interpret ionic conductivity. To explore structural patterns beyond prediction, we applied dimensionality reduction and clustering by employing t-SNE and DBSCAN, as well as Self-Organizing Maps (SOM), which reveal distinct membranes clusters, some of which were enriched with high anion conductivity. The optimized graph convolutional network (GCN) achieves an R² score of 0.94 on the test set. While the descriptor-based model is slightly less accurate, it offers valuable insights through SHAP analysis, highlighting electronic, topological, and compositional descriptors as key drivers of conductivity. Additionally, atom-level saliency maps from GCN provide spatial explanations for conductive behavior, revealing the importance of polarizable and flexible regions. This work contributes to the accelerated and data-driven design of high-performance AEMs.

This publication has 30 references indexed in Scilit: