Application of artificial neural networks on a series of anilides molecules based DFT calculations.

Abstract

For a quantitative structure-activity relationship of activity (pI50) inhibiting the reduction of DCIP photosynthetic, a series of 49 selected molecules anilides was modeled using the software GaussView (03) using the DFT method with the B3LYP function and the 6-31G basis (d) the descriptions obtained are purely electronic.The whole constitutes the inhibitory activity and the calculated electronic descriptors were statistically processed with the principal component analysis (PCA), multiple linear regression (MLR), non-linear multiple regression (MNLR) and artificial neural network (ANN). The results obtained by the artificial neural network (ANN) show that the planned activities are in good agreement with the experimental results with equal correlation coefficient R = 0,87.To determine the architecture of the network, we varied the number of hidden layers, the number of neurons in the hidden layer, the transfer functions and the pairs of transfer functions. The best results were obtained with network architecture [7-3-1], the activation functions (tansig-Purelin) and a learning algorithm Levenberg-Marquardt