Combining descriptive and predictive modeling to systematically design depth filtration‐based harvest processes for biologics
Peter Liu,Michael Hartmann,Ajay Shankaran,Hong Li,John Welsh
DOI: https://doi.org/10.1002/bit.28765
IF: 3.8
2024-06-05
Biotechnology and Bioengineering
Abstract:Because mechanistic understanding of depth filtration is limited, we developed a hybrid modeling workflow to predict the harvest process performance of mammalian cell culture, solely from filter and feedstream attributes. This hybrid approach enables prospective experimental design, reduces total experimental load, and accelerates development under strict timelines. Advances in upstream production of biologics–particularly intensified fed‐batch processes beyond 10% cell solids–have severely strained harvest operations, especially depth filtration. Bioreactors containing high amounts of cell debris (more than 40% particles <10 μm in diameter) are increasingly common and drive the need for more robust depth filtration processes, while accelerated timelines emphasize the need for predictive tools to accelerate development. Both needs are constrained by the current limited mechanistic understanding of the harvest filter‐feedstream system. Historically, process development relied on screening scale‐down depth filter devices and conditions to define throughput before fouling, indicated by increasing differential pressure and/or particle breakthrough (measured via turbidity). This approach is straightforward, but resource‐intensive, and its results are inherently limited by the variability of the feedstream. Semi‐empirical models have been developed from first principles to describe various mechanisms of filter fouling, that is, pore constriction, pore blocking, and/or surface deposit. Fitting these models to experimental data can assist in identifying the dominant fouling mechanism. Still, this approach sees limited application to guide process development, as it is descriptive, not predictive. To address this gap, we developed a hybrid modeling approach. Leveraging historical bench scale filtration process data, we built a partial least squares regression model to predict particle breakthrough from filter and feedstream attributes, and leveraged the model to demonstrate prediction of filter performance a priori. The fouling models are used to interpret and provide physical meaning to these computational models. This hybrid approach–combining the mechanistic insights of fouling models and the predictive capability of computational models–was used to establish a robust platform strategy for depth filtration of Chinese hamster ovary cell cultures. As new data continues to teach the computational models, in silico tools will become an essential part of harvest process development by enabling prospective experimental design, reducing total experimental load, and accelerating development under strict timelines.
biotechnology & applied microbiology