Rede neural para previsão de velocidade de produção enzimática da ampicilina

Abstract

The antibiotics correspond to the largest production network of the pharmaceutical industry used in human and veterinary medicine, mainly in poultry and swine farming. The enzymatic production of antibiotics is complex and non-linear. Therefore, its modelling owns rate equations with a large number of parameters. In this context, neural networks emerge as an alternative to conventional kinetic models. It is a computational tool that presents mathematical models inspired by human neurons, which allows, after training, to assist in the study and improvement of production data. The application of neural networks in biotechnological processes has been used for years, in order to estimate state variables, to determine biological phases, and to monitor biochemical reactors. The present study utilized the neural network methodology to obtain a kinetic model for the enzymatic synthesis of ampicillin from phenylglycine methyl ester and 6-aminopenicillanic acid catalyzed by penicillin G acylase, immobilized on chitosan. The methodology for the training and simulation of the neural network used in this study began with obtaining a database of antibiotic production rates (Ampicillin), hydrolysis of the reagent and the product. The database was assembled in pairs of concentration and velocity data to form a matrix of training, simulation and validation. Commands in Silab were used to shape the neural network. The number of hidden layers, the number of neurons and the mathematical model were defined for the training, simulation and validation of the neural network. Different transfer functions were tested and the Pureling function proved to be the less precise. Multiple training sessions were accomplished, changing its parameters, in order to obtain the best combination to predict the rate of antibiotic production. The transfer function Tansig, with 100 neurons in the hidden layer, provided the configuration that obtained the best results, in regards to the errors.