A research paper by Prof. Koichiro Kato, Prof. Tsuyohiko Fujigaya, and former Q-ENERGY Fellowship student Phua Yin Kan at Kyushu University has been published in the international journal Journal of Materials Chemistry A, issued by the Royal Society of Chemistry.

The research team developed a Human-in-the-loop framework that integrates explainable artificial intelligence (XAI), ChatGPT, and expert knowledge to design anion exchange membrane (AEM) materials, which are key components in fuel cells and water electrolysis systems.

Although artificial neural networks (ANNs) can accurately predict polymer properties, their decision-making processes are often difficult for researchers to interpret, making them “black-box” AI models. In this work, the researchers introduced a novel dimensionality reduction strategy that enabled the application of explainable AI techniques to high-dimensional ANN models.

Using this framework, the team successfully extracted quantitative molecular design guidelines that can be directly utilized by experimental researchers, including the effectiveness of biphenyl structures and the importance of side-chain lengths with eight-bond distances.

The results are expected to reduce trial-and-error in materials development and accelerate screening for promising candidate materials. Furthermore, the database and source code developed in this study have been released on GitHub, enabling broader applications beyond AEM materials to other functional polymer systems.

Publication Information

Journal
Journal of Materials Chemistry A

Title
“Orchestrating Explainable AI, ChatGPT, and Human Expertise: A Framework for Extracting Polymer Design Guidelines”

Authors
Phua Yin Kan, Nana Terasoba, Manabu Tanaka, Tsuyohiko Fujigaya, Koichiro Kato

DOI
https://doi.org/10.1039/D5TA06120B

Kyushu University Press Release
https://www.kyushu-u.ac.jp/ja/researches/view/1473