Repository logo
 

Data intelligence for process performance prediction in biologics manufacturing

Accepted version
Peer-reviewed

No Thumbnail Available

Type

Article

Change log

Authors

Sewell, D 
Turner, R 
Field, R 

Abstract

Despite the availability of large amount of data in bioprocess databases, little has been done for its retrospective analysis for process improvement. Historic bioprocess data is multivariate time-series, and due to its inherent nature, is incompatible with a variety of statistical methods employed in data analysis resulting in the lack of a tailored methodology. We present here an integrative framework of knowledge discovery tailored for handling historical bioprocess datasets. The pipeline successfully predicts process performance at harvest from an early time point, and robustly identifies the most relevant process parameters to model process performance. We present the utility of this pipeline on biologics manufacturing data from upstream bioprocess development for antibody production by mammalian cells. The proposed multi-model system that employs machine learning can predict performance at harvest after two weeks of operation with satisfactory accuracy employing data generated as early as on the sixth day of the culture.

Description

Keywords

Biologics manufacturing, Culture performance prediction, Data mining, Time-series analysis, Machine learning, Two-dimensional modelling

Journal Title

Computers and Chemical Engineering

Conference Name

Journal ISSN

0098-1354
1873-4375

Volume Title

146

Publisher

Elsevier BV

Rights

All rights reserved
Sponsorship
European Commission (289126)
Biotechnology and Biological Sciences Research Council (BB/K011138/1)
Medimmune (AstraZeneca)