Materials discovery through interpretation: An application to proton-conducting oxides

Machine learning has become a pivotal tool in materials discovery, but conventional data-driven models often behave as “black boxes” limited by the scope of their training data. In this perspective, we outline the current status of machine learning-driven materials discovery and present Materials Discovery through Interpretation (MDI), a framework that integrates scientific reasoning into the ML loop to enhance the reliability and impact of materials design. Instead of relying solely on statistical correlations, the MDI iteratively incorporates domain knowledge and insights from experimental and/or computational results such as phase formability and materials functionality into the predictive process. This interpretive strategy enables meaningful extrapolation even when underlying physics and chemical correlations remain poorly understood, guiding the identification of new candidate materials while simultaneously refining the models. Applied to proton-conducting oxides, MDI successfully identified previously unrecognized compounds, including those achieving conductivity above 0.01 Scm −1 at 300 °C, a critical benchmark for fuel cell electrolytes, while providing clear rationales for their selection. By combining predictive power with interpretability, MDI offers a flexible pathway for accelerating materials innovation beyond the confines of existing data. • Proposes MDI framework integrating scientific reasoning to ML for materials design. • MDI reliably predicts material functions using domain knowledge and expert insights. • Applied to proton conductors, MDI finds stable oxides with >0.01 S cm −1 at 300 °C.