For several years we have been systematically synthesizing and crystallizing families of related compounds, providing insights into polymorphism, similarity and crystal structure formation. These families can get rather large, which has necessitated developing computerized approaches to searching for patterns and packing similarities. Whilst many of the properties under investigation can be driven by conventional or primary intermolecular interactions there is an increasing awareness that weak or secondary interactions (in addition of course to shape and steric factors) can have considerable influence on the arrangement of the solid state. Accordingly we use two approaches that are not reliant purely on characterizing and understanding primary, hydrogen bonded, interactions.<br/>Quantification of similarity in terms of dimensionality is important in understanding polymorphism, phase transitions and crystal growth. The XPac program disregards the standard descriptors of intermolecular interactions and is a geometrical analysis that assigns vectors between equivalent points in adjacent molecules in order to try to capture the effects of diffuse Van der Waals interactions. Pairs, or indeed large families, of vector representations can then be used to index the degree of similarity between all the members of a collection. This has the advantage of not only being agnostic to intermolecular interactions but also to an extent molecular similarity – it is indeed possible to compare oranges and lemons!<br/>Our second method takes quite the opposite approach. Using molecular, interaction and topological statistical descriptors gleaned from all possible characteristics of a crystal structure (we have used around 4000!) it is possible to build statistical models. Again when looking at related sets of structures, one can compare models and also look for correlations between descriptors. In this way it is not only possible to characterize similarity, but also relate physical properties to structural characteristics and answer questions such as “will it crystallize?” or “what coformers will produce a cocrystal?”.<br/>We are lacking tools to mine really big and diverse collections in this way – our methods will be illustrated through studies on our own compound/structure libraries, however we demonstrate the potential, and shortcomings, of mining much larger databases by these approaches.

Crystallographic informatics: Similarity and statistics

251st National Meeting (2016)

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