Application of a Recurrent Neural Network to Prediction of Drug Dissolution Profiles

The Elman Recurrent Neural Network was employed for the prediction of

in-vitro dissolution profiles of matrix controlled release theophylline pellet preparation, leading to the potential use of an intelligent learning system in the development of pharmaceutical products with desired drug release characteristics. A total of six different formulations containing various matrix ratios of substance to control the release rate of theophylline were used for experimentation. By using the leave-one-out cross-validation approach, the dissolution profiles of all the matrix ratios were consumed for training, except for one set that was taken as a reference profile, with which the network predicted profiles were compared. Performance of the network was assessed using the similarity factor, f ₂ , a criterion for dissolution profile comparison recommended by the United States Food and Drug Administration. Simulation results indicated that the Elman network was capable of predicting dissolution profiles that were similar to the reference profiles with an error of less than 8%. In addition, the Bootstrap method was used to estimate the confidence intervals of the f ₂ values. The results revealed the potential of a neural-network-based intelligent system in solving non-linear time-series prediction problems in pharmaceutical product development.