The clinical and commercial success of cell-derived proteins has generated significant research interest in the optimisation of their manufacturing processes. A focal point has been the upstream cultivation of mammalian cells for recombinant glycoprotein production, such as monoclonal antibodies (mAbs) (Oliveira et al., 2015), with optimisation options involving genetic/host engineering, media/feed formulation, and operating strategy. At a cellular level, such optimisation is carried out by harnessing the understanding of the genomic-phenotypic relationship. Despite many decades of scientific development, such efforts can often be time-consuming and resourceful expensive. In response to the increased development of optimisation strategies and manufacturing control, in silico models have been developed in the form of mathematical representations of physiological processes, in a quantifiable extension of controlled experimental conditions. They are employed to facilitate in-depth understanding of the underlying cellular physiology, be it in organising generated information from omics analysis or gaining insight about component interactions in complex systems.
Nevertheless, mechanistic models, such as kinetic cell culture models, are typically rigid in their structure and contain parameters with set values, such as Monod constants for growth rate calculation and yields of metabolic pathways. The present thesis aims to explore hybrid modelling formulations, involving stoichiometric and kinetic models to create an adaptive modelling framework for cell metabolism. The principal motivation for developing the methodology was to formulate a segregated model informed by the stoichiometric metabolic model that could be fitted to a particular dataset and validated on experimental datasets from a series of platform process combinations.
Additionally, machine learning algorithms were integrated with metabolic data analysis using stoichiometric models to identify extracellular and intracellular metabolic patterns that could correlate with productivity and growth. These can, in turn, be used to support cell line screening and selection protocols. Finally, the hybrid methodology has also been used to define feature-class decision boundaries for the system at hand, which can guide future process optimisation research via the identification of cellular constraints.