Novel soft-sensor applications and mechanistic models for biomanufacturing with Escherichia coli

Abstract

This thesis investigates model-based tools for developing soft sensors and control systems in bioprocessing applications involving Escherichia coli. The approaches integrate kinetic modeling and nonlinear uncertainty propagation with existing soft-sensor strategies based on first-principle elemental balances, enhancing robustness amid dynamically changing substrate concentrations.A novel extension of the soft sensor involves real-time estimation of substrate uptake capacity qSmax and yield coefficient YX/S. Combining superimposed pulse-feeds with a nonlinear particle-filter (PF) observer allows simultaneous estimation of both parameters, contingent on the practical observability under specific processing conditions.Incorporating the prediction of recombinant protein production, a mechanistic model is developed based on bioreactor cultivations of Escherichia coli producing lactate dehydrogenase (LDH). The model distinguishes soluble protein formation from insoluble inclusion body (IB) formation. Through analysis, selection, and parameter identification, a cost-to-go model predictive controller is devised. This unique concept involves offline dynamic programming optimization for obtaining optimal control input trajectories before the process starts. During the process, the cost-to-go matrix is utilized with a model predictive controller (MPC), enabling global optimization of the control input beyond the prediction horizon, leading to improved product yields in simulation studies compared to a standard MPC approach.The PF soft-sensor concept is extended from upstream cultivation to the downstream refolding step of IBs back into their native form. Utilizing online intrinsic fluorescence measurements of Trp and Tyr residues, the soft sensor predicts the total refolding reaction rate and distinguishes folding states into native protein and protein aggregates. This concept, applied to IB refolding processes with limited online monitoring capabilities, provides deeper insights into refolding kinetics.