Standardized Chemical Structures and Descriptors

Chemical structures were standardized with the function StandardiseMolecules from the R package camb [41 ] with the following options: (i) inorganic molecules were removed, and (ii) molecules were selected irrespectively of the number of fluorines, chlorines, bromines or iodines present in their structure, or of their molecular mass. Morgan fingerprints [42 (link),43 (link)] were calculated using RDkit (release version 2013.03.02) [44 ,45 ]. For the calculation of unhashed Morgan fingerprints [45 ], each compound substructure in the dataset, with a maximal diameter of four bonds, was assigned to an unambiguous identifier. Subsequently, substructures were mapped into an unhashed (keyed) array of counts. Physicochemical descriptors (PaDEL) [46 (link)] were calculated with the function GeneratePadelDescriptors from the R package camb. The R package vegan was used to generate the distributions of pairwise compound similarities (Jaccard distance) [47 ].
The amino acids composing the binding site of the mammalian cyclooxygenases considered in this study (Table 1), were described with five amino acid extended principal property scales (5 z-scales) [48 (link)]. Z-scales were calculated with the R package camb [41 ].