We developed a procedure that identifies "novel" biologically active compounds that are expected to be distinct from known active compounds with respect to specificity and other such characteristics. The procedure involves mapping a set of known active compounds (training set) onto all possible hydrogen bonding and lipophilic interaction sites of an enzyme active site and flagging those interactions that are utilized by the training set. These flagged sites are removed (except for those that are deemed critical binding sites), leaving only potential interaction sites not utilized by the active compounds. Once unflagged sites were enumerated, pharmacophore models were then generated, scored, and prioritized where the top pharmacophore model was used to search 3D databases for identifying new leads. This procedure was applied to HIV-1 protease inhibitors. Several compounds retrieved by the top pharmacophore model were identified as moderately active (in mMolar range).
pharmacophores; structure-based design; 3D databses
