[5Yin-A] Poster 8

[5Yin-A-08] Interpretable Modeling of Organic Materials Synthesis Using Computational Chemistry Descriptors and Machine Learning

〇Kiho Matsubara¹, Ryohei Kakuchi¹, Takeshi Matsuda², Kei Takahashi³, Noriaki Seko⁴, Masaaki Omichi⁴ (1. Gunma Univ., 2. Hannan Univ., 3. Osaka Metropolitan Univ., 4. QST-Takasaki)

Keywords:

machine learning,Materials Chemistry

Recently, the importance of data-driven approaches has increased in the field of chemistry, particularly in materials synthesis. However, in organic materials synthesis, the large number of interacting factors makes systematic understanding highly challenging. In this study, we address such complex systems by constructing a machine learning (ML) model that uses molecular descriptors calculated via computational chemistry as input features, enabling the prediction of reactivity and analysis of feature importance in materials synthesis. The target reaction is radiation-induced graft polymerization (RIGP), a representative organic materials synthesis process in which solvent effects, grafting monomers, and substrate properties interact in a highly complex manner. In particular, quantitative prediction of the influence of solvent selection on reactivity has remained an unresolved issue. Therefore, we conducted synthesis experiments using a wide variety of solvents to obtain reactivity data, and corresponding molecular properties were calculated by computational chemistry and used for ML. As a result, we demonstrate that the reactivity of previously unseen solvents can be explained based on molecular properties rather than simple correlations.