Improved electron-conformational method of pharmacophore identification and bioactivity prediction. Application to angiotensin converting enzyme inhibitors.

The electron-conformational (EC) method of pharmacophore (Pha) identification and bioactivity prediction, suggested earlier, is given here two major improvements. First, an atomic index of orbital and charge controlled interaction is introduced to better represent the ligand (substrate) in its interaction with the bioreceptor. Second, the multiconformational problem is considered in view of ligand-receptor binding states, resulting in essential simplification of the expression of bioactivity. The details of the improved EC method are demonstrated in application to the problem of angiotensin converting enzyme (ACE) inhibitors. The Pha of the latter is identified by separation of the heavily populated conformations of the chosen 51 compounds (the training set), calculation of the electronic structure, construction of their EC matrixes of congruity, and processing of the latter in comparison with the activities to reveal a common submatrix of all the active only compounds that describes the Pha. The latter contains three oxygen atoms plus a fourth atom X = S, N, O at certain interatomic distances and with restricted electronic parameters (within assumed tolerances), the position of the atom X being more changeable from one active compound to another. For quantitative prediction of the bioactivity, an expression is deduced which takes into account the duly parametrized influence of auxiliary groups (AG) which, being positioned outside the Pha, either diminish the activity (antipharmacophore shielding) or enhance it. It is shown that in case of many conformations of the same compound only one of them, that of the lowest energy which has the Pha, should be parametrized. The 15 parameters chosen to represent the AG in case of ACE inhibitors are weighted by variational (adjustable) coefficients which are determined from a regression treatment of the calculated versus known activities in the training set. Then the formulas with known coefficients are used to validate the method by calculating the bioactivity of other compounds not used in the training set. The prediction of the activity proved to be more than 90% (within experimental error and available compounds) qualitatively (yes, no) and about 60%-70% quantitatively.