One of the primary challenges during process characterization is establishing the suitability of the small scales models used to generate the data. The traditional approach to qualification of scale-down models has been to verify that lab-scale results match large-scale results to within an arbitrary precision. An alternate approach is to directly incorporate scaling offsets into predictive models. However, obtaining a precise estimate of scaling offsets can be difficult when large-scale datasets are relatively small. As such, it is essential to use a statistical method that can account for this uncertainty. This talk will explore the use of Bayesian modeling to account for scaling offsets in predictive models. Bayesian methods are well-suited to this task because they inherently account for uncertainty, while offering a flexible platform for the implementation of hierarchal models. This talk will explore the use of Bayesian statistics for both upstream and downstream applications. In a downstream application, Bayesian statistics were used to account for scale-offsets in a high-throughput chromatography system using miniature chromatography columns. In the past, concerns about the scalability of high-throughput methods have been a barrier to using high-throughput data in process characterization models. Using Bayesian techniques, chromatography data from high-throughput and traditional lab-scale DOE studies were combined with manufacturing-scale data to generate models accounting for scaling-offsets. In an upstream application, a hierarchal Bayesian model was used to generate a predictive model to account for scaling offsets in bioreactor performance parameters. Bioreactor scale-up is not a simple linear process, and a hierarchal Bayesian approach enables an assessment of how the effects of input process parameters may differ across scales. Overall, the Bayesian approach is a flexible method that facilitates more informed predictions of large-scale performance by enabling the incorporation of scaling offsets and prior knowledge into predictive models.

Using a Bayesian framework to account for scale-down model offsets during process characterization

255th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.