Integrating Data-Driven Strategies into Model Predictive Control for Enhanced Production of Human Interferon α2b in Glycoengineered Pichia pastoris

Abstract

The biopharmaceutical industry is experiencing rapid growth, necessitating scalable optimization and control strategies to meet strict process objectives. Model predictive control (MPC) offers a robust framework for regulating complex bioprocesses; however, its performance critically depends on the availability of reliable process models. While mechanistic models are often preferred, practical limitations have accelerated the adoption of data-driven approaches. In this study, we evaluate the applicability of artificial neural networks (ANNs) and Gaussian process (GP) models in MPC for fed-batch cultivation of glycoengineered Pichia pastoris to produce human interferon α2b (huIFNα2b). Experiments were performed in a fermentation calorimeter with real-time monitoring of P. pastoris metabolism through metabolic heat rate, capacitance, and exhaust gas analysis. Comparative results demonstrate that GP-based MPC achieved superior process control, efficient substrate utilization, and a 1.1-fold increase in huIFNα2b productivity relative to ANN-based MPC. Furthermore, GP-based adaptation of feeding strategies reduced methanol consumption by 14% compared with ANN-based control. These findings highlight the potential of GP-driven MPC as a promising tool for enhancing productivity and sustainability in industrial bioprocesses.