Compilation of the Co-crystal Database

Statistical analysis requires a large number of observations, each of which is described by a set of well-defined attributes. The presence of a co-crystal structure in the CSD serves as an observation in our case. The structure is evidence that the constituent molecules can form a co-crystal with each other. Molecular descriptors, similar to those used in QSAR/QSPR studies, were calculated for pairs of co-crystallising molecules. These paired molecular descriptors provided the numerical attributes for statistical analysis. 

The first step in the generation of the database of molecular properties in cocrystals was the retrieval of co-crystal structures from the CSD. Co-crystals were defined as organic structures containing at least two eharge neutral nonsolvent molecules with different structural formulas. To avoid possible statistical bias by the frequent occurrence of some eommonly used eo-formers e.g., urea, 4,4 -bipyridine, see Table 1 in ref. 7), struetures with these moleeules were excluded. Initial survey of the data showed that inelusion eompounds, which consist of a large and a small molecule, form a distinet group in the data set. Co-crystals of molecules with more than 30 and with less than six nonhydrogen atoms were thus removed and the analysis foeused on eo-erystals other than inclusion compounds. The final data set eontained 710 struetures. 

For each molecule in these struetures, 131 molecular descriptors were calculated. These included simple atom, bond and group eounts, hydrogen bond donor and acceptor counts, topological indices, a variety of partitioned and charge weighted surface areas and deseriptors of molecular size, shape, polarity 

---