Abstract
Materials informatics (MI), which emerges from the integration of materialsscience and data science, is expected to greatly streamline the materialdiscovery and development. The data used for MI are obtained from bothcomputational and experimental studies, while their integration remainschallenging. In our previous study, we reported the integration of thesedatasets by applying a machine learning model that captures trends hidden inthe experimental datasets to compositional data stored in the computationaldatabase. In this study, we use the obtained data to construct materials maps,which visualize the relation in the structural features of materials, aiming tosupport study by the experimental researchers. The map is constructed using theMatDeepLearn (MDL) framework, which implements the graph-based representationof material structures, deep learning, and dimensional reduction for the mapconstruction. We evaluate the obtained materials maps through statisticalanalysis and found that the MDL using message passing neural network (MPNN)enables efficient extraction of features that reflect the structural complexityof materials. Moreover, we found that this advantage does not necessarilytranslate into improved accuracy in predicting material properties. Weattribute this unexpected outcome to the high learning performance inherent inMPNN, which can contribute to the structuring of data points within thematerials map.