Prediction of pK(a) values of neutral and alkaline drugs with particle swarm optimization algorithm and artificial neural network

Abstract

A prediction model of pKa values of neutral and alkaline drugs based on particle swarm optimization algorithm and back propagation artificial neural network, called PSO–BP ANN, was established. PSO–BP ANN model was proposed using back propagation artificial neural network trained by particle swarm optimization algorithm, and used to predict the pKa values. The five parameters, including relative N atom number, Randic index (order 3), relative negative charge, relative negative charge surface area and maximum atomic net charge, were selected by particle swarm optimization algorithm and used as input variables of the model. The output variable in the proposed model was pKa values. The experimental results showed that the three layers (5–7–1) prediction model had a good prediction performance. The absolute mean relative error, root mean square error of prediction and square correlation coefficient were 0.5728, 0.0512 and 0.9169, respectively. The pKa values of neutral and alkaline drugs were positively correlated with the value of maximum atomic net charge, but the pKa value decreased with the increase in the other four parameters.