Classification of some active compounds and their inactive analogues using two three-dimensional molecular descriptors derived from computation of three-dimensional convex hulls for structures theoretically generated for them

Two three-dimensional (3D) molecular descriptors are used to classify 73 protease inhibitors against the human immunodeficiency virus:type 1 (HIV-1). X-ray structures of these HIV-1 protease bound inhibitors are used as templates to generate the most probable bioactive conformations of the inhibitors. A convex hull computation algorithm is applied to each structure generated. The frequency of atoms lying on the vertexes of each hull is counted. Vertexes of the same atomic charge state are then gathered together as a set of commonly exposed groups for all the structures generated. The first 3D descriptor is computed as the maximum molecular path length among any three distinct commonly exposed groups, while the second 3D one is computed as the maximum molecular path length among any three atoms of nonconvex hull vertexes. We find that the 73 HIV-1 protease inhibitors can be classified by the first 3D descriptor into two groups, which agrees with the result of visual classification using the activity data as a criterion for these compounds. The classification scheme is then used to classify a database of 427 active trypsin inhibitors and their inactive analogues. The structures of these:compounds are generated theoretically from steps of energy minimization and molecular dynamics. Classification for all these compounds is performed using the SYBYL hierarchical clustering method on the first 3D descriptor and then the second 3D one computed. It is found that some inactive analogues are completely separated from the active inhibitors at the first stage of classification using the first 3D descriptor. Most of the highly active inhibitors are classified into a cluster at the second stage of classification using the second 3D descriptor. Finally, most of these highly active inhibitors are separated from all the accompanying inactive analogues in the cluster through a structural alignment process using a set of commonly exposed groups determined for them.