Accelerated materials property predictions and design using motif-based fingerprints

Data-driven approaches are particularly useful for computational materials\ndiscovery and design as they can be used for rapidly screening over a very\nlarge number of materials, thus suggesting lead candidates for further in-depth\ninvestigations. A central challenge of such approaches is to develop a\nnumerical representation, often referred to as a fingerprint, of the materials.\nInspired by recent developments in chem-informatics, we propose a class of\nhierarchical motif-based topological fingerprints for materials composed of\nelements such as C, O, H, N, F, etc., whose coordination preferences are well\nunderstood. We show that these fingerprints, when representing either molecules\nor crystals, may be effectively mapped onto a variety of properties using a\nsimilarity-based learning model and hence can be used to predict relevant\nproperties of a material, given that its fingerprint can be defined. Two simple\nprocedures are introduced to demonstrate that the learning model can be\ninverted to identify the desired fingerprints and then, to reconstruct\nmolecules which possess a set of targeted properties.\n