Prediction of Dissolution Data Integrated in Tablet Database Using Four-Layered Artificial Neural Networks.

Abstract

A large number of dissolution data were measured and integrated into a previously constructed tablet database composed of 14 kinds of compounds as model active pharmaceutical ingredients (APIs) with contents ranging from 10 to 80%. The database has contained physicochemical and powder properties of APIs, together with basic physical attributes of tablets such as the tensile strength and the disintegration time. In order to enhance the value of this database, drug dissolution data are essential to improving key information for designing tablet formulations. A four-layered artificial neural network (4LNN), newly implemented in commercially available software, was employed to predict dissolution data from physicochemical and powder properties of APIs. Our results showed that an excellent model for the prediction of dissolution data was achieved with 4LNN method. The function of 4LNN was appreciably better than that of conventional three-layered model, despite both models adopting the same number of nodes and algorithms for activation functions. Furthermore, linear regression models resulted in poor prediction of dissolution data.