Soft Sensing of Intracellular States for CHO Cell Bioprocessing with Ensemble Kalman Filters

Abstract

In biotherapeutic manufacturing, product quality such as glycosylation profile is typically assessed only after harvest, limiting opportunities for corrective action during cell culture operation. Intracellular nucleotide sugar donors (NSD) directly determine glycosylation outcomes but are rarely measured, even offline, due to analytical complexity and process disruption. As a result, quality-related decisions remain constrained to fixed operating strategies. This work introduces a model-based soft sensing framework to infer NSD concentrations from readily available extracellular measurements. A Bayesian state estimation approach based on the Ensemble Kalman Filter (EnKF) is developed to reconstruct unmeasured intracellular states during CHO cell culture. An imperfect kinetic process model is combined with noisy extracellular measurements, explicitly accounting for process variability and measurement uncertainty through ensemble-based propagation and updates. The framework is validated using four independent experiments with distinct feeding perturbations that are not used for model calibration. Although the open-loop model exhibited substantial mismatch for both extracellular metabolites and intracellular NSDs, EnKF assimilation of extracellular measurements corrected key metabolic profiles. Building on these corrected extracellular dynamics, the EnKF demonstrated robust estimation of a growth-determining amino acid, asparagine, from correlated extracellular states. Based on the improved extracellular and amino acid estimates, the framework further enabled reliable inference of intracellular NSDs across all experiments.