Advancing bioprocessing with ensemble kalman filter: from state estimation to knowledge transfer
File(s)
Author(s)
Yu, Luxi
Type
Thesis
Abstract
The biopharmaceutical industry is undergoing a rapid digital transformation, driven by the adoption of computational tools that enhance process understanding and enable accurate prediction, monitoring, and control. Yet, real-time automation remains challenging due to limited model transferability, process variability, and the lack of reliable analytics, all of which underscore the need for robust state estimation methods.
This work introduces the Ensemble Kalman Filter (EnKF) for the first time in the bioprocessing literature, addressing challenges in state estimation and model adaptability. First, EnKF is applied as a soft sensor for intracellular nucleotide sugar donors (NSD), which serve as metabolic precursors for glycosylation, a critical quality attribute (CQA) of recombinant glycoproteins. While mechanistic models describe NSD synthesis, they are system-specific and poorly suited to uncertainty quantification. By combining extracellular data with a mechanistic model, EnKF enables model inference of NSD concentrations without direct measurements. Experimental validation confirms the feasibility of this soft sensing strategy. These inferred states are then used as inputs for glycan prediction through Recurrent Neural Network (RNN), supporting real-time product quality control.
The second part of this thesis applies EnKF for dual state and parameter estimation to adapt mechanistic models across bioprocess systems. Traditionally, these models require extensive reparameterization to transfer between bioreactor scales, cell lines, or process conditions. Here, EnKF continuously assimilates real-time data to update both states and parameters, enabling model adaptation and revealing biologically meaningful insights through the evolving parameter ensemble. Such dynamics may, for example, indicate shifts in metabolic pathways or substrate utilization unique to a particular cell line. It represents a flexible and reliable approach that can be generalized to any system or scale.
This work introduces the Ensemble Kalman Filter (EnKF) for the first time in the bioprocessing literature, addressing challenges in state estimation and model adaptability. First, EnKF is applied as a soft sensor for intracellular nucleotide sugar donors (NSD), which serve as metabolic precursors for glycosylation, a critical quality attribute (CQA) of recombinant glycoproteins. While mechanistic models describe NSD synthesis, they are system-specific and poorly suited to uncertainty quantification. By combining extracellular data with a mechanistic model, EnKF enables model inference of NSD concentrations without direct measurements. Experimental validation confirms the feasibility of this soft sensing strategy. These inferred states are then used as inputs for glycan prediction through Recurrent Neural Network (RNN), supporting real-time product quality control.
The second part of this thesis applies EnKF for dual state and parameter estimation to adapt mechanistic models across bioprocess systems. Traditionally, these models require extensive reparameterization to transfer between bioreactor scales, cell lines, or process conditions. Here, EnKF continuously assimilates real-time data to update both states and parameters, enabling model adaptation and revealing biologically meaningful insights through the evolving parameter ensemble. Such dynamics may, for example, indicate shifts in metabolic pathways or substrate utilization unique to a particular cell line. It represents a flexible and reliable approach that can be generalized to any system or scale.
Version
Open Access
Date Issued
2025-06-05
Date Awarded
2026-02-01
Copyright Statement
Attribution-Non Commercial-No Derivatives 4.0 International Licence (CC BY-NC-ND)
Advisor
Kontoravdi, Cleo
del Rio Chanona, Antonio
Publisher Department
Department of Chemical Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)