In February I had the opportunity to share some thoughts on the Opportunities and Challenges in Biologics Manufacturing Process Data Analytics Innovation at the National Academy of Sciences. Below is a video of the 20-minute presentation.
The talk is summarized on page 6 of the proceedings of the meeting:
Jack Prior, head of Manufacturing Science, Global Manufacturing Science and Technology at Sanofi, provided his view of the current state of process data analytics. He stated that collecting, managing, and analyzing data are becoming more challenging as the industry moves from simply describing what is happening to predicting and controlling what is going to happen. He continued that there are barriers to leveraging process data to improve operations—physical barriers that involve capabilities to measure, access, and organize data and organizational barriers that involve questions of trust and the desire to analyze and act on the data. He noted that trust means not only accepting that data are accurate but proving that they are robust enough to use for a particular application. He then described three realities of data analytics for biological processes. First, manufacturing data are not the same as data from a designed experiment, and one has to be careful not to equate correlation with causality. Furthermore, the sources of variation are often not contained in the data. Second, biological processes are nonlinear and time-variant, and the industry needs to move away from multivariate analyses. Third, although the industry typically does not generate “big data”, integration and analysis of industry data require investment.
Prior presented 24 technologies in the manufacturing innovation lifecycle and noted that where a technology falls in the innovation lifecycle for a specific company depends on which technologies the company needs to enable successful production of its products. He highlighted three technologies on the horizon: digital twins that encapsulate process knowledge in a real-time “twin„ that monitors and predicts process behavior, mixing validation that uses computational fluid dynamics to examine the entire vessel and surpasses conventional mixing assessment, and data-lake systems that store all the raw data that can then be used for various applications. He continued by noting several barriers in various phases that are inhibiting innovation. In the exploration phase, one needs data engineers, modelers, and domain experts who have access to large quantities of the right data that can be matched to the right problem. In the industrialization phase, there needs to be a critical-mass market for vendor commercialization and a digital infrastructure that allows integration and validation. For initial filings, one needs the right fit for a critical product or process and enough lead time to build the innovation into early process development; a problem raised earlier in the workshop is the fear of being first with the risk of approval delays. In the commercialization phase, barriers include substantive investment in legacy platforms and the need for global regulatory acceptance. To accelerate digital innovation, Prior concluded, the industry needs to invest in digital infrastructure and integration, there needs to be a joint effort to advance agile software validation, and FDA innovation-promotion programs and efforts in global regulatory harmonization need to continueNational Academies of Sciences, Engineering, and Medicine. 2020. Innovations in Pharmaceutical Manufacturing: Proceedings of a Workshop—in Brief. Washington, DC: The National Academies Press. https://doi.org/10.17226/25814.