Computation screening for incorrectly determined cocrystal structures

Pharmaceutical solid forms, like salts and cocrystals, play a crucial role in drug formulation. Despite differing mainly by a single hydrogen atom, the regulatory requirements set by the US Food and Drug Administration for these forms vary significantly. We previously developed a DFT-based computational method to distinguish salts from cocrystals. This method, validated on 95 structures, performed well for systems where hydrogen bonds were longer than 2.613 (16) Å. Here, benefits of the rSCAN functional over the PBE functional are discussed. We expand the dataset to 404 cocrystal models. Analysis confirms that 301 of these forms are indeed cocrystals. Additionally, 87 salt-cocrystal continuum forms are identified and 16 cocrystals are classified as possible salts. These 16 problematic structures are further investigated and for seven of them, single crystals were grown and their structure determined using single-crystal X-ray diffraction. Among the phases exhibiting salt-like behaviour, five of them are identified as salts. In some cases, rSCAN alone gives unreliable results for strong hydrogen bonds, but these discrepancies are often corrected using better-renormalized or hybrid functionals (i.e. r2SCAN, PBE0 and PBE50). For future calculations, we recommend using the r2SCAN functional for salt-cocrystal differentiation, as it provides reliable results for O-H...N bonds longer than 2.554 (5) Å. The r2SCAN functional offers a good balance between accuracy and computational efficiency for systems with longer O-H...N bonds.