In extension and improvement of previous results, a novel method is worked out for pharmacophore identification and activity prediction in structure-activity relationships. In this method, as in our previous works, each molecular system (conformation) of the training set is described by a matrix with both electron structural parameters (atomic charges, bond orders, etc.) and interatomic distances as matrix elements. This description includes a rather full geometry of charge and/or reactivity distribution thus providing a much better representation of the molecular properties in their interaction with the target. By multiple comparison of these matrices for the active and inactive compounds of the training set, a relatively small number of matrix elements are revealed that are common for all the active compounds and are not present in the same combination in the inactive ones. In this way a set of electronic and geometry parameters is obtained that characterize the pharmacophore (Pha). A major improvement of this scheme is reached by introducing the anti-pharmacophore shielding (APS) and a proper treatment of the conformational problem. The APS is defined as molecular groups and competing charges outside the basic skeleton (the Pha plus the inert neighbor atoms that do not affect the activity) that hinder the proper docking of the Pha with the bioreceptor thus diminishing (partially or completely) the activity. A simple empirical formula is derived to estimate the relative contribution of APS numerically. Two main issues are most affected by the APS: (1) the procedure of Pha identification is essentially simplified because only a small number of molecular systems with the highest activity and simplest structures (systems without APS) should be tried for this purpose; (2) with the APS known numerically, we can make a quantitative (or semiquantitative) prediction of relative activities. The contributions of different conformations (of the same molecular system) that possess the Pha and different APS is taken into account by means of a Boltzmann distribution at given temperatures. Applied to an example, rice blast activity, this approach proved to be rather robust and efficient. In validation of the method, the screening of 39 new compounds yields approximately 100% (within experimental error) prediction probability of the activity qualitatively (yes, no), and with r(2) = 0.66 quantitatively.
