ARTIFICIAL NEURAL NETWORKS APPLIED TO THE QUANTITATIVE STRUCTURE-ACTIVITY RELATIONSHIP STUDY OF PARASUBSTITUTED PHENOLS

The artificial neural network (ANN) model with back-propagation of err or is used to study the quantitative structure-activity relationship o f para-substituted phenol derivatives between the biological activity and the physicochemical property parameters. Network parameters are op timized, and an empirical rule for dynamically adjusting the network's learning rate is proposed to improve the network's performance. The r esults show that the three-layer ANN model gives satisfactory performa nce, with f(x)=1/(1+exp(-x)) as the network node's input-output transf ormation function and the number of hidden nodes 10. The network gives the mean square error (mse) of 0.036 when predicting the biological a ctivity of 26 para-substituted phenol derivatives. This result compare s favourably with that obtained by the conventional methods.