Optimization of preparation of Insulin nanoparticles composed of quaternized aromatic derivatives of chitosan using artificial neural networks

Shadab Shahsavari¹ Gita Bagheri² Reza Mahjub³ Farid Abedin Dorkoosh⁴

1) Chemical Engineering Department, Varamin-Pishva Branch, Islamic Azad University, Varamin, Iran.
2) Chemical Engineering Department, Shahr-e-Qods Branch, Islamic Azad University, Tehran, Iran.
3) Department of Pharmaceutics, Faculty of Pharmacy, Hamedan University of Medical Sciences, Tehran, Iran
4) Department of Pharmaceutics, Faculty of Pharmacy, Tehran University of Medical Sciences, Tehran, Iran

محل انتشار : کنفرانس بین المللی علوم و مهندسی(icesconf.com)
Abstract :
The aim of this research was to develop an artificial neural network (ANN) in order to design a nanoparticulate oral drug delivery system for insulin. The pH of polymer solution (X1), concentration ratio of polymer/insulin (X2) and polymer type (X3) are considered as the input values and the particle size, zeta potential, PdI, and entrapment efficiency (EE%) as output data. ANNs are employed to generate the best model to determining the relationships between input and response values. In this research, a multi-layer percepteron with different topologies has been tested in order to define the one with the best accuracy and performance. The optimization was used by minimizing the error between the predicted and observed values. Three training algorithms (Levenberg –Marquardt (LM), Bayesian- Regularization (BR), and Gradient Descent (GD) were employed to train ANNs with various numbers of nodes, hidden layers and transfer functions by random selection. The accuracy of prediction data were assayed by the mean squared error (MSE).The ability of all algorithms was in the order: BR> LM> GD. Thus, BR was selected as the best algorithm.
Keywords : Artifical neural network (ANN), Levenberg- marquadt algorithm, Bayesian Regulation (BR) algorithm, Gardient Descent algorithm, Insulin nanoparticle