Analysis of known crystals to design polymorph prediction strategies.

Computational analysis of known crystals was used to provide insight into polymorph prediction strategies. It was hypothesized that intermolecular crystalline interactions are inadequately accounted for in current computational strategies. The QEq method of partial charge assignments was applied to the crystal milieu (CMc) and to the isolated molecule (IMc). For 90 known crystal structures, these two partial charging methods showed that the unit cell energy was almost always lower (more stable) with CMc partial charge assignments. In simulations on a model drug, the CMc charging re-ranked the unit cell energies of possible polymorphs in a way that aided the identification of physically reasonable packings more easily than IMc. Decomposing the unit cell energy for known crystals showed that the cohesive IMc deviation from CMc was dominant. On the other hand, for simulations, the corresponding IMc conformational component of unit cell energy dominates when conformational adaptive strategies (CA) are used; this was not apparent with rigid-body simulations (RB). Furthermore, evidence is presented that polymorph strategies that use unit cell energy for optimizing are especially prone to this undesired conformational bias. Suggestions are made for polymorph strategies that might enhance the probability of identifying physically relevant polymorphs in the future.