A grand challenge in the area of computational chemistry is the prediction of crystal structures from first principles. The area of crystal structure prediction (CSP) has seen important advances in the past few years, both in the underlying theoretical models and algorithms and applications in, for example, pharmaceutical solid form screening and the discovery of functional materials. As we develop these methods, it is also crucial to develop rules for interpreting their results. For example, what is the relevant energetic range for observable polymorphism or what is the typical energetic driving force for co-crystallization? To answer these questions, we make use of the wealth of information in the Cambridge Structural Database, which we combine with large scale simulations. This presentation focusses on co-crystallization and the information that can be learned from lattice energy calculations on a large set of observed structures. The study investigates the range of co-crystallization driving force for observed co-crystals: the stability of a co-crystal with respect to the separate crystallization of its pure components and the structural features that lead to stable co-crystals. We also present attempts to apply machine learning to predict the co-crystallization driving force from molecular features alone. In this work, we apply crystal structure prediction to supplement the dataset to which machine learning can be applied.

Energetics of co-crystal formation: Informing prediction through combining the database with large scale simulations and machine learning

2019 Fall National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.