Prediction of Electrophoretic Mobilities of Organic Acids Using Artificial Neural Networks with Three Different Training Functions

Abstract

The quantitative structure–mobility relationship study has been done to develop the correlation between the electrophoretic mobility of a number of common organic acids in capillary electrophoresis and their molecular structures. Molecular descriptors calculated from structure, were used to represent molecular structures. A subset of the calculated descriptors was used in model development. Multiple linear regression and artificial neural networks are utilized to construct the linear and non-linear prediction models. This paper focuses on investigating the role of weight update function in artificial neural networks. Therefore, artificial neural networks with three different weight update functions including Levenberg–Marquardt algorithm, gradient descent with variable learning rate back propagation and resilient back propagation were trained. Finally, obtained results using three different artificial neural networks have been compared.