Digital Chemical Engineering (Jun 2023)
A scoping review of supervised learning modelling and data-driven optimisation in monoclonal antibody process development
Abstract
Background: Supervised learning modelling and data-driven optimisation (SLDO) methods have only recently gathered interest in the monoclonal antibody (mAb) platform process development application, but have already demonstrated their advantages over traditional approaches in reducing development costs and accelerating research efforts. With potential usage in multiple unit operations, there is a need for mapping existing SLDO methodologies with the corresponding mAb applications. Methods: We performed a scoping review of mAb process development studies with at least one SLDO method published prior to April 26, 2022. A team of four independent reviewers conducted a search and synthesised characteristics of the eligible studies from four literature databases. Results: We identified 30 relevant studies from 1785 citations and 118 full-text papers. 70% were upstream studies (n = 21), and the majority of papers were published between 2010 and 2022 (n = 27, 90%). Multivariate data analysis (MVDA) techniques were identified as the most common SLDO methods (n = 11), and were typically used to model heterogeneous and high-dimensional bioprocess data. While the main usage of SLDO in process development was predictive modelling, a few studies also focused on data pre-processing, knowledge transfer, and optimisation. Conclusions: Despite the data challenges inherent to the mAb industry, SLDO has been demonstrated to be an efficient solution to some process development use cases such as knowledge transfer, process characterisation, optimisation, and predictive modelling. As biopharmaceutical companies are advancing their digital transformation, SLDO methods will need to be further developed and studied from a more integrative perspective to remain competitive against other platform development approaches.
