Systematic development of predictive mathematical models for animal cell cultures

Abstract Fed-batch cultures are used in producing monoclonal antibodies industrially. Existing protocols are developed empirically. Model-based tools aiming to improve productivity are useful with model reliability and computational demand being important. Herein, a systematic framework for developing predictive models is presented comprising of model development, global sensitivity analysis, optimal experimental design for parameter estimation, and predictive capability checking. Its efficacy and validity are demonstrated using a fed-batch structured/unstructured model of antibody-secreting hybridoma cultures. Global sensitivity analysis is first used to identify sensitive model parameters (initial values estimated from batch cultures). Information-rich data from an optimally designed fed-batch experiment are then used to estimate these parameters, resulting in good agreement between simulation and experimental results. Finally, the model's predictive capability is confirmed by comparison with an independent set of fed-batch cultures. This approach systematises the process of developing predictive cell culture models at a minimum experimental cost, enabling model-based control and optimisation.