Recurrent neural network modeling applied to expanded bed adsorption chromatography of chitosanases produced by Paenibacillus ehimensis

Abstract

Nonlinear autoregressive networks with external input (NARX) and multilayer perceptron (MLP) has been used to predict the activity and protein content for flow-through, washing and elution steps during expanded bed adsorption chromatography of chitosanases produced by Paenibacillus ehimensis. Bed expansion as well as the influence of particulate-containing feedstock study showed a stable bed operation without significant impact on the yield as well as purification factor when the cells were fed to the column. Also, NARXs showed a better performance to predict the chitosanolytic activity as well as total protein than MLPs.