Supporting the implementation of chemometrics models in bioprocess from development to commercial stage
Authors:
- Thibault HelleputteTHThibault Helleputte
- Thomas CornetTCThomas Cornet
- Ahmed KanfoudAKAhmed Kanfoud
- Thomas Des TouchesTTThomas Des Touches
- Pascal GerkensPGPascal Gerkens
- Patrick DumasPDPatrick Dumas
- Antonio Gaetano CardilloACAntonio Gaetano Cardillo
- Gaël de LannoyGLGaël de Lannoy
Preprint
Abstract
Manufacturing bioproducts involves using living organisms to produce a substance of interest. Cell culture is a crucial yet highly variable step. Monitoring, predicting, or even controlling the course of this growth requires data. For multiple reasons, the perspective of developing Process Analytical Technology based on chemometrics, rather than proceeding to actual samplings from the bioreactor, is strongly appealing and could lead to full digital twins of bioreactors. This paper addresses multiple roadblocks encountered along this modeling approach: proper model performance evaluation, limited amounts of data at process development stages, up-scaling between development and commercial stages, and regulatory concerns. These challenges are addressed by combining spectrometry data and offline readouts, simple and advanced machine learning techniques, and a good knowledge of the biomanufacturing activity and constraints. This paper uses vaccine data from multiple actual products at different scales to illustrate the considered developments. Authors contributions: Thibault Helleputte, Thomas Cornet, Ahmed Kanfoud, Thomas des Touches, Pascal Gerkens and Gaël de Lannoy were involved in the conception and design of the study and/or the development of the study protocol. Pascal Gerkens, Patrick Dumas, Antonio Gaetano Cardillo and Gaël de Lannoy participated in the acquisition of data. Thibault Helleputte, Thomas Cornet, Ahmed Kanfoud and Thomas des Touches implemented the analyses, applied them to the data, and interpreted the results. All authors were involved in drafting the manuscript or revising it critically for important intellectual content. All authors had full access to the data and approved the manuscript before it was submitted by the corresponding author.Keywords
This publication has 11 references indexed in Scilit:
- Feature selection with prior knowledge improves interpretability of chemometrics modelsChemometrics and Intelligent Laboratory Systems, 2023
- Practical data-driven modeling and robust predictive control of mammalian cell fed-batch processComputers & Chemical Engineering, 2023
- The Shape of Learning Curves: A ReviewIEEE Transactions on Pattern Analysis and Machine Intelligence, 2022
- RMTL: an R library for multi-task learningBioinformatics, 2018
- Model-Based Methods in the Biopharmaceutical Process LifecyclePharmaceutical Research, 2017
- Inferring statistically significant features from random forestsNeurocomputing, 2015
- Feature Selection by Transfer Learning with Linear Regularized Models2009
- A notion of task relatedness yielding provable multiple-task learning guaranteesMachine Learning, 2008
- Random ForestsMachine Learning, 2001
- Multitask LearningMachine Learning, 1997