Predicting Compositional Changes of Organic-Inorganic Hybrid Materials with Augmented CycleGAN

Stoichiometric information, despite its simplicity, has demonstrated excellent predictive power over a range of materials properties.¹ Efficient sampling methods are necessary to search this chemical space for target properties (inverse design). A promising candidate for this task is generative models that can produce synthetic samples by learning the distribution of real samples.<br/>In this study, we describe a way to generate compositions of hybrid organic-inorganic crystals through adapting Augmented CycleGAN,² a novel generative model that can learn many-to-many relations between two domains. Specifically, we investigate the composition change upon amine swap: For a specific chemical system (set of elements) crystalized with amine A, how would chemical compositions change if it is crystalized with amine B? By training with limited data from Cambridge Structural Database, our model can generate realistic chemical compositions for hybrid crystalline materials. This model can also utilize abundant unpaired data (compositions of different chemical systems), a feature that traditional supervised methods lack. The generated compositions can be used for many tasks, for example, as input fed to a classifier that predicts structural dimensionality.<br/><b>References</b><br/>(1) Jha, D.; Ward, L.; Paul, A.; Liao, W.; Choudhary, A.; Wolverton, C.; Agrawal, A. ElemNet: Deep Learning the Chemistry of Materials from Only Elemental Composition. <i>Sci Rep</i> <b>2018</b>, <i>8</i> (1), 17593.<br/>(2) Almahairi, A.; Rajeswar, S.; Sordoni, A.; Bachman, P.; Courville, A. Augmented CycleGAN: Learning Many-to-Many Mappings from Unpaired Data. <i>arXiv:1802.10151 [cs]</i> <b>2018</b>.