Esterification modification of graphene oxide for Metronidazole release: Optimization and modelling using response surface methodology and artificial neural Networks

Abstract

The esterification modification of graphene oxide was performed by the reaction with the terephthalic acid and allyl alcohol in the presence of p-toluene sulfonic acid as a catalyst for metronidazole delivery. The artificial neural network was used for properly modelling and predicting the percentage of metronidazole sorption. The artificial neural network model with the topology of 3:6:1 (three input variables, six neurons in two and three hidden layers, and one output variable) using feed-forward back-propagation learning was developed. The influence of various parameters, e.g., pH, contact time, and initial drug concentration on the sorption process was investigated through central composite design using response surface methodology at five levels. The maximum adsorption efficiency of the drug was observed at 94.26 % under optimal conditions. The Levenberg-Marquardt algorithm performed better than the scaled conjugate gradient algorithm. Layer 2 with 6 neurons with the least mean square error with 3 epochs was chosen as the optimal layer and neuron for drug sorption. The experimental results were examined by using linear and non-linear forms of the isotherms and kinetics models. According to the coefficient of determination and sum of squares of errors, it was shown that the non-linear forms of Langmuir isotherm and pseudo-2nd-order kinetic models are best fit the experimental results. The thermodynamic factors indicated that the drug sorption on adsorbent is endothermic and spontaneous. The in-vitro release study showed that the drug release in simulated gastric fluid (pH = 1.2) was significantly lower compared to the release in simulated intestinal media (pH = 7.4). At pH = 1.2, 27.60 % of the drug has been released over 1 h compared to 98.97 % in simulated intestinal media over 6 h. The fitting of kinetic results by the pharmacokinetic models shows the non-Fickian diffusion from the nanocarrier.