Prediction of dynamic plasmid production by recombinant escherichia coli fed-batch cultivations with a generalized regression neural network

A generalized regression neural network with external feedback was used to predict plasmid production in a fed-batch cultivation of recombinant Escherichia coli. The neu ral network was built out of the experimental data obtained on a few cultivations, of which the general strategy was based on an initial batch phase followed by an exponen tial feeding phase. The different cultivation conditions used resulted in significant differ ences in bacterial growth and plasmid production. The obtained model allows estimation of the experimental outputs (biomass, glucose, acetate and plasmid) based on the bioreactor starting conditions and the following on-line inputs: feeding rate, dissolved oxygen concentration and bioreactor stirring speed. Therefore, the proposed methodol ogy presents a quick, simple and reliable way to perform on-line feedback prediction of the dynamic behaviour of the complex plasmid production process, based on simple on-line input data obtained directly from the bioreactor control unit and with few cultiva tion experiments for neural network learning.