This paper proposes an interpretable deep learning architecture incorporating the attention mechanism to predict material properties and understand structure-property relationships. The architecture's predictive capabilities are validated using various datasets, showing comparable performance to state-of-the-art models. Furthermore, analysis indicates the importance of atomic local structures in interpreting structure-property relationships for properties like molecular orbital energies and formation energies. The architecture's ability to identify crucial structural features accelerates material design.

Tien-Sinh Vu, Minh-Quyet Ha, Duong-Nguyen Nguyen, Viet-Cuong Nguyen, Yukihiro Abe, Truyen Tran, Huan Tran, Hiori Kino, Takashi Miyake, Koji Tsuda, Hieu-Chi Dam