Formulation of controlled release acyclovir loaded chitosan nanoparticles was optimized based on the optimization
technique using response surface method (RSM) and artificial neural network (ANN) simultaneously to develop a
model to identify relationships between variables affecting drug nanoparticles. In this research, the goal was to create a
representation of three irregular factors, i.e. concentration of acyclovir, concentration ratio of chitosan/ Tripolyphosphate
(TPP) and pH on response variables. ANN was used to create a fit model of formulations via these four training algorithms
including: Levenberg–Marquardt (LM), Gradient Descent (GD), Bayesian–Regularization (BR) and BFGS Quasi-
Newton (BFG) were applied to train ANN containing a various hidden layer, applying the testable data as the training set.
Criterion to stop training was the divergence of the RMSE (root mean squared error) between target and output values.
Both methods including gradient descent and Levenberg-marquardt have showed similar results in the data formulation.
Corresponding to batch back propagation (BBP)-ANN performance, a gain in pH of polymer solution reduced the size
and polydispersity index (PdI) of nanoparticles. Moreover, decreases in the concentration ratio of chitosan/TPP consequently
cause an increase in entrapment efficiency (%EE).
The next aim of this research was to investigate the performance of predictive algorithms. For this reason each training
algorithm in order to consider the accuracy of predictive ability was evaluated and the result was as follow: LM > BFGs >
GD > BR.
Keywords: Artificial neural network (ANN), Response surface methodology (RSM), Backpropagation, Training algorithms,
Drug delivery, Acyclovir.
Rights & PermissionsPrintExport