QSARs for a series of inhibitory anilides

Neural network was applied to study a series of anilide herbicides inhibiting photosystem II (PSII). The molecules were encoded by a set of dimensional parameters. The model established using a back-propagation algorithm of neural network (r = 0.967; N = 76) was superior to that obtained using multiple linear regression (r = 0.922; N = 76). The descriptorʹs contributions to the PS II inhibitory activity pI50 were calculated by a method which consists in analysing the statistical coefficients between observed and calculated pI50 using neural network, when a descriptor is taken off. The results obtained were interpreted in terms of interactions molecule-receptor site.