loading page

Data-driven digital twin models for forecasting multi-step ahead profiles of mammalian cell culture performance
  • +2
  • Dong-Yup Lee,
  • Seo-Young Park,
  • Sun-Jong Kim,
  • Cheol-Hwan Park,
  • Jiyong Kim
Dong-Yup Lee
Sungkyunkwan University School of Chemical Engineering

Corresponding Author:[email protected]

Author Profile
Seo-Young Park
Sungkyunkwan University School of Chemical Engineering
Author Profile
Sun-Jong Kim
Sungkyunkwan University School of Chemical Engineering
Author Profile
Cheol-Hwan Park
Sungkyunkwan University School of Chemical Engineering
Author Profile
Jiyong Kim
Sungkyunkwan University School of Chemical Engineering
Author Profile

Abstract

Recently, enormous culture profiles and datasets from biomanufacturing processes to produce recombinant therapeutic proteins (RTP) such as monoclonal antibodies (mAbs) could be generated by virtue of the advancement in process analytical techniques and artificial intelligence (AI). Thus, now it is highly necessary to develop AI-based data-driven models (DDMs) and exploit them accordingly in order to further enhance operational efficiency and accelerate reliable product supply. Since bioprocess is a complex and dynamic system, DDMs are practical and particularly useful to describe the intrinsic relationship among biological and process parameters and cell culture conditions by capturing inherent patterns and to produce high-quality RTP under consistent operations as well as to decrease cost and time by predicting incipient or abrupt faults during the cell cultures. In this work, we provide the practical guideline for choosing the best DDM on given mAb-producing Chinese hamster ovary (CHO) cell culture data sets, enabling us to forecast culture performance such as VCD, and mAb titer as well as glucose, lactate and ammonia concentrations in real time manner. Via the case study with 32 fed-batch data sets of CHO cell cultures, we suggested best combination of model elements including AI algorithms and multi-step ahead forecasting strategies, for good prediction in terms of the computational load as well as the model accuracy and reliability, which is applicable to implementation of interactive data-driven model within bioprocess digital twins. We believe this systematic study can help bioprocess engineers to start developing predictive DDMs with their own data and learn how their cell cultures behave in near future, thereby making proactive decision possible.
02 Dec 2022Submitted to Biotechnology and Bioengineering
02 Dec 2022Submission Checks Completed
02 Dec 2022Assigned to Editor
02 Dec 2022Review(s) Completed, Editorial Evaluation Pending
03 Dec 2022Reviewer(s) Assigned
10 Jan 2023Editorial Decision: Revise Major
14 Mar 20231st Revision Received
14 Mar 2023Assigned to Editor
14 Mar 2023Submission Checks Completed
14 Mar 2023Review(s) Completed, Editorial Evaluation Pending
21 Mar 2023Reviewer(s) Assigned
03 Apr 2023Editorial Decision: Revise Minor
04 Apr 20232nd Revision Received
10 Apr 2023Assigned to Editor
10 Apr 2023Submission Checks Completed
10 Apr 2023Review(s) Completed, Editorial Evaluation Pending
10 Apr 2023Editorial Decision: Accept