A commercial-scale biopharmaceutical manufacturing process typically operates as a batch process that includes multiple batch processing unit operations such as scale-up bioreactors, harvest and purification. The goal of a process monitoring system in biopharmaceutical manufacturing is to ensure that a higher degree of reproducibilty across campaigns is achieved. This is a critical prerequisite to ensure that the quality targets are met and the product conforms to its specifications.. In process industries, batch process monitoring tools provide valuable real-time information to the process engineers and plant operators not only to effectively monitor and control their batch processes, but also to support their decision making. Despite over two decades of research and advancements in industrial batch process monitoring and control, biopharmaceutical processes are still grappling with the small data problem. In fact, it is common in biopharmaceutical manufacturing to have only a few engineering batches for a product with no commercial production history at a manufacturing facility before the actual good manufacturing practice (GMP) campaign. While limiting the number of engineering runs improves speed to market and reduces manufacturing costs, it also limits the amount of representative data available for establishing a monitoring baseline, which presents several problems. First, with small data sets it is nontrivial to capture the inherent process variability under the normal operating conditions (NOC) in its entirety. Second, most of the multivariate monitoring methods, such as principal component analysis (PCA) and partial-least squares (PLS) are big-data technologies that are effective only when there are large historical data sets.<br/><br/>In this talk we will discuss several machine-learning methods that allow us to learn in small data scenarios. Based on this new system understanding, we generate virtual campaign data under normal operations on-demand using non-parametric modeling methods. Using data from the virtual campaigns, standard batch monitoring techniques are used to effectively monitor a biopharmaceutical manufacturing process. Encouraging results are reported from practice.<br/>

Learning from small data in the era of big data for real-time biopharmaceutical process monitoring

253rd National Meeting (2017)

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