Predicting molecular crystal growth

Abstract

The incorporation of crystal growth processes extends the utility of predictive techniques beyond simply crystal geometry and a thermodynamic ranking with the ultimate goal being actionable information to target specific crystallisation. Pairwise model interaction energies with implicit solvation have been previously demonstrated in the CrystalClear protocol to provide a computationally efficient route to incorporate such processes, though some limitations restricted its use. We present several advancements to overcome these limitations, incorporating a native implementation of the semi-empirical GFN2-xTB model with SMD solvation, and validating both GFN2-xTB and the CE-1p model against reference lattice energies and solubilities. Further, automated enumeration of symmetry-unique faces and construction of solvated surface energies as part of the same protocol facilitates direct prediction of solvent-dependent equilibrium crystal shapes while eliminating the tedium and errors frequently associated with manual construction of surface geometries required to evaluate surface free energies. To demonstrate the potential of these advancements we present two case studies, oxindole and ritonavir, evaluating the accuracy of the crystal growth protocol in reproducing experimental results for particle size effects, polymorph stability and more. All methods presented here are freely available and implemented in open source software, ensuring accessibility, extendability and transferability to the broader community.