Forecasting for fermentation operational decision making

Abstract

in Wiley InterScience (www.interscience.wiley.com).An awareness of the likely future behavior of a batch or a fed-batch fermentation process is valuable information that can be exploited to improve product consistency and maximize profitability. For example, by making operational policy changes in a feedforward control sense, improved consistency can be facilitated, while prior knowledge of batch productivity, or the end time, can help determine the downstream processing configuration and upstream process scheduling. In this article, forecasting methods based on multivariate batch statistical data analysis procedures are contrasted with case-based reasoning (CBR). Additionally, the importance of appropriate statistical data prescreening and the choice of a suitable metric for case selection are investigated. Two industrial case studies are considered, a fedbatch pharmaceutical fermentation and a batch beer fermentation process. It is demonstrated that, following appropriate statistical prescreening of the data, in terms of forecasting performance, CBR is comparable to linear projection to latent structures (PLS), for the more straightforward problem, i.e., the batch beer fermentation, while for the more complex case-the pharmaceutical process-CBR exhibits enhanced performance over PLS.

Progress toward forecasting product quality and quantity of mammalian cell culture processes by performance‐based modeling

“…With commonly applied statistical methods, including the techniques used in this publication, it is not possible to correlate a data matrix with a single value. Unfolding was done by arranging the off‐line data as described in the literature (Figure A). As a result of the unfolding a vector with t ‐times k elements was obtained for each batch.…”

“…For example, PLS‐R was applied to predict the time‐point at which the optimal alcohol content was reached in a brewery process. However, the same statistical approach was not successful when applied to a more complex antibiotics process . Other authors succeeded in forecasting the final product concentration such as Ignova et al for a penicillin G, Lennox et al for an unspecified product and Gunther et al for a recombinant protein …”

Bis‐arylidene Oxindoles as Anti‐Breast‐Cancer Agents Acting via the Estrogen Receptor

“…Kennedy and Krouse [21]), since they can be used to extract relations in a data set – either historical process data or data resulting from a dedicated statistical design of experiments – without requiring detailed knowledge of any underlying mechanism. Especially in the case of process data, appropriate data pretreatment is essential before using the data in the frame of data‐driven or mechanistic model building [22]. Once developed, a fermentation process model can ideally predict performance of future production facilities considering the scale of operation, the process design and process conditions.…”

Predictive Monitoring of Shake Flask Cultures with Online Estimated Growth Models

“…Fermentation processes, which depend on the behavior of microorganisms and their interaction with the substrate [ 1 ], are often prone to variability. Despite being invaluable prerequisites for process optimization, only a few publications are devoted to forecasting of bioprocess signals [ 16 , 17 , 18 ]. Shake flask cultures deserve particular attention in this context.…”