Co-crystal Screening Methods

However, grinding in the presence of a small amount of a liquid phase (liquid-assisted grinding or LAG) was found to be highly efficient for the synthesis of co-crystals, resulting in rapid and quantitative formation of cocrystals, typically within 20 minutes. A suitable illustration of the greater efficiency of LAG for co-crystal screening is the proof-of-principle study of the 

In addition to efficiency and speed, LAG also provides a highly crystalline product. As a result, co-crystal formation is readily observed through X-ray powder diffraction, which can also be used to characterise the new materials structurallyIndeed, Karki and co-workers have recently demonstrated that the reaction speed, high crystallinity and quantitative yield associated with LAG co-crystallisation enable the rapid structural characterisation of new cocrystals involving simple components, such as the model API theobromine, within 24 hours. More recently, it was demonstrated how products of mechanochemical co-crystal formation can also be structurally characterised through crystal structure prediction methods, followed by comparison of measured and calculated X-ray powder diffraction patterns. 

The co-crystal database does not contain information about pairs of molecules that do not form co-crystals with each other, so common machine learning techniques cannot be applied to derive a predictive model, for example, by fitting a scoring function. Instead, we based our selection criteria on descriptor differences between co-crystal formers in the CSD. Four of the descriptors, SjL, MjL, FNO and dipole, were chosen for further use. The other descriptors in Table 5.1 contain redundant information and show wider distributions than those selected. 

a simple classification scheme was tested. A pair of molecules was classified as likely to form a co-crystal if the difference between the molecules in all four descriptors is smaller than a predefined cut-off value. The 9th deciles of the differences in the CSD (Table 5.1) were used as cut-offs, that is, 90% of known co-crystals show a difference smaller than the cut-off values. The scheme was applied to the 218 screening experiments that had been performed previously (see above). Effectively, this test probes whether selecting co-formers using shape and polarity in addition to supramolecular synthons gives a benefit. 

The molecular models for the calculation of the descriptors were not taken from the co-crystals, since these would not be known prior to performing the experiments. They were extracted from the single-component crystals of the test compounds when available and they were calculated using molecular mechanics (MMFF94 force field) otherwise. 

---