PRODUCT FORMATION KINETICS IN A RECOMBINANT PROTEIN PRODUCTION PROCESS

Protein formation in recombinant protein production cannot yet be modeled in a way sufficiently accurate for process supervision and control. Here we propose using a new hybrid approach based on mass balances for the state variables involved, where the kinetics are represented by artificial neural networks (ANN). We first demonstrate by means of simulations that this method works well even when the networks are trained on noisy process data. Then, secondly, we show that the method is applicable to real fermentation data. As an accompanying example we use an E.coli culture that produces a recombinant protein, namely the green fluorescent protein GFP, which remains dissolved within the cytoplasm. For this case the ANN resulted in a concrete relationship between the specific product formation rate π, the specific growth rate μ and the specific product concentration p/x. The π(μ)-part of the relationship confirms what was obtained with a conventional approach and the additional information about the influence of the specific product concentration characterizes the metabolic load of the cell.