Crossed Mixture Design and Artificial Neural Networks: An Efficient Approach to Cell Culture Medium Optimization

Abstract

Although many commercially available cell culture media exist, none of them are able to optimally meet the specific requirements of every cell line used for large-scale recombinant protein production. Through a novel approach to develop a medium for culturing genetically engineered mammalian cells, the optimal blends of six compounds that should be present in culture media used in recombinant protein production were determined. The aim of this work was to define the composition of two different serum-free culture media by testing two groups of compounds that are added to a basal formulation, through a crossed mixture design. The goals pursued were to maximize the quantity of active secreting cells the productivity and the quality of the secreted molecule (in terms of glycosylation) while minimizing toxic accumulation of catabolites during the culture, using both batch and continuous processes. Empirical data obtained from crossed mixture design were used to train artificial neural networks for each response. Two artificial neural networks were selected for each response and used to predict the responses for 800 new combinations of H1, H2, H3, E1, E2, and E3. These predicted responses were combined to calculate a Global Desirability Function (D). The combinations of the six components which originated the highest values of function D were chosen to be tested in a continuous process in a 5 L-perfused bioreactor.