ABSTRACT Anion exchange membranes (AEMs) are key components in emerging energy technologies, yet their development is hindered by the challenge of simultaneously achieving high hydroxide conductivity and durable alkaline stability. These properties are governed by multiscale, coupled effects of polymer architecture, microphase separation, and operating conditions, making AEM exploration slow and largely empirical. Here, we propose SPARK, a structure‐property graph attention network with prior knowledge of chemistry embedding, to accelerate AEM molecular design through robust candidate prioritization and mechanism‐relevant interpretability. SPARK embeds chemical prior knowledge into molecular graphs and employs a dual‐channel architecture that separately encodes hydrophilic ionic and hydrophobic non‐ionic segments, explicitly capturing microphase separation and ion‐transport channel formation. It translates AEM structures into five‐level performance grades for hydroxide conductivity and alkaline stability with high accuracy, outperforming conventional machine‐learning baselines. Attention‐based interpretation further pinpoints structural units associated with ion transport and alkaline degradation, providing actionable guidance to mitigate the conductivity‐stability trade‐off. Finally, SPARK is validated by pre‐grading to‐be‐synthesized AEM candidates with good agreement to experiments, and the accompanying software package is publicly released to facilitate broader adoption and data‐driven AEM design.
Chen et al. (Thu,) studied this question.