Abstract
Creation of well-curated databases tailored to specific structural motifs can underpin and drive materials discovery, as the properties of materials are governed by composition and structure. The role of such motifs in directing the intricate interplay between structure and properties is exemplified by intermetallic compounds with structures that contain kagome layers that exhibit a variety of exotic physical states. Two prevailing approaches have previously been applied to identify such materials: evaluation of structural topology or geometry assessment, however, both present limitations if deployed individually. We augment topological screening with geometrical filtering to allow versatile control over the identification of kagome layers. Applying this approach with minimal further constraints labels over 9000 kagome-containing intermetallics which are assigned to four structural classes, revealing connections between symmetry, composition, direct space structure, and flatband electronic structures in reciprocal space. A machine learning model is used to predict new element combinations that favour the formation of kagome layers. Several highly-ranked phase fields correspond to known kagome-containing materials that were absent from the training dataset, demonstrating that the workflow can identify chemistries affording kagome layers. This motivates the extension of the approach beyond kagome to other property-conferring motifs, such as honeycomb, square planar or triangular plane nets.