Integrating data-driven strategies into model predictive control for enhanced production of human interferon α2b in glycoengineered Pichia pastoris

The biopharmaceutical industry is witnessing exponential growth and continuously seeking for a scalable optimisation and control strategy to meet the process constraints and objectives. Model predictive control (MPC) has emerged as a robust approach to realise enhanced control over process parameters in bioprocesses. The effectiveness of MPC hinges on the availability of a robust and workable process model. Although mechanistic models are preferred, practical constraints may limit their feasibility and researchers have resorted to data-driven approaches. In this work, we demonstrate the applicability of two data-driven models namely Artificial Neural Networks (ANN) and Gaussian Process (GP) in MPC applications in fed-batch cultivation of glycoengineered P. pastoris for the production of human interferon α2b (huIFN α2b). The experimental verification was carried out by performing fed-batch cultivation in a fermentation calorimeter. Real time monitoring of P. pastoris metabolism was facilitated by metabolic heat rate, capacitance, and exhaust gas analyser measurements. GP-based MPC demonstrated better control, efficient utilization of substrate and 1.1-fold enhanced huIFNα2b productivity compared to ANN-based MPC. Moreover, optimal feeding adapted by GP-based MPC resulted in a 14% decrease in methanol utilization compared with ANN-based MPC.