Fast Molecular Crystal Structure Prediction Using Sampling by Analogy to Previously Predicted Landscapes

We demonstrate a proof of concept for an approach to fast molecular crystal structure prediction using analogy to hypothetical crystal structures of related molecules. Our method constructs different valid analogues of predicted crystal structures of previously studied molecules and uses these analogues as trial structures in crystal structure prediction workflows for new similar molecules, facilitating successful structure predictions with greatly reduced sampling compared to a conventional CSP search. Early testing on a set of six different families of similar rigid molecules demonstrates that our method can predict both known crystal structures and low lying structures on the target landscape as it would be determined by more complete conventional structure predictions. We show that our approach does so with greater success than quasi-random methods with equally minimal sampling and additionally that trial structures generated via our analogue-based approach lie closer to local minima than structures generated via quasi-random sampling, leading to further reductions in cost. Through these investigations, we show the promise of the proof of concept, which has potential applications where quick structure predictions are needed, such as high throughput studies or preliminary testing.