Boehringer Ingelheim and the quest for the golden batch
- ChallengeUnderstand variation in biological batch processes
- SolutionsEmploy the PI System to facilitate multivariate data analysis techniques and attribute data
- BenefitsReduce production variance, generate greater batch potency, reduce batch destruction, and increase yield
The story of pharmaceutical production is often a quest for the golden batch: a repeatable process that consistently optimizes yield and quality. Recently, an animal health subsidiary of Boehringer Ingelheim, a top international pharmaceutical company, began collaborating with Sartorius Stedim Data Analytics. The joint project uses multivariate data analysis techniques (MVDA) to create golden-batch trajectories based on historic data. These trajectories can then be used for real-time monitoring and prescriptive process control for higher-potency batches.
Compliance is always a chief concern for biopharmaceuticals. Results must fall within a proven acceptable range (PAR) of acceptable variability. “It’s about demonstrating that your batch is under control—that’s the compliance aspect of this,” said Will A. Penland, principal data scientist for animal health at Boehringer Ingelheim. “The government takes that very seriously. You need to demonstrate that your process is under control and that you are capable of meeting the specification limits.” Minimizing batch variation and understanding the causes of variability is a crucial but difficult task. “When you are dependent on biological processes for the creation of your product, there is a lot of inherent variability that you have to contend with,” Penland added. “With a biofermentation process, you can put the same thing in, and if you have production variances, you can get something quite different day by day.” The problem is that variation can arise at many points in the process. Part of the mystery is always which process attributes or conditions are responsible for the greatest degree of variation. The sheer number of potential sources of variation makes this a critical but difficult question to answer when pursuing the golden batch.
Penland and his team began by asking themselves a simple question: “What if?” What if they could reduce production variation? What if they could identify crucial process attributes? And then, what if they could target those key process components to reduce production variance, generate greater batch potency, reduce batch destruction, and increase yield? The answer seemed to lie with using MVDA and advanced statistical-modeling techniques like principle-component analysis (PCA) and partial least squares (PLS). But implementing MVDA presented another set of obstacles. Gaining a better understanding of process variability meant monitoring the process at many levels. This complex monitoring generates many kinds of data. In order to use the company’s data-analytic software, SIMCA, it needed a way to first aggregate and contextualize the data from these potential sources of variation.
Managing data is the hard part
“Eighty to 95% of the work is managing your data,” Penland explained. This is where the PI System came to Boehringer Ingelheim’s rescue. Using Asset Framework, the company created data tags of different classes for various types of data streams related to the fermentation process. The PI System allowed it to bring its diverse data streams together, aligning its SCADA, LIMS, process control, and attribute data, with its time-resolved spectroscopic data. Boehringer Ingelheim also triggered process steps within its programmable logic controller and used a step numbering system to create Event Frames. It now uses a parent-child event-frame model to keep a close eye on the various stages of the fermentation process, particularly the growth phase. The parent event represents the batch, and the child events are the various steps within the fermentation process. Templates enable the company to ensure consistent configuration of its process stages. Templates are crucial for a large international company like Boehringer Ingelheim. “The analytics is one thing, but the challenge these days is to make these things scalable, reproducible, and easy to template to move across platforms,” said Chris McCready, lead data scientist at Sartorius Stedim Data Analytics. “If we come to a customer and they have PI with a good configuration, then we can have a monitoring process, advanced analytics—the whole thing—up and running in a matter of days. And if they have Asset Framework, then we can take what we did in New York and we can apply it in Singapore. It’s very cut-and-paste.” Now that the PI System enables Boehringer Ingelheim’s advanced data analytics and batch-evolution modeling, the company is becoming more data driven, predictive, and proactive. It can now see things as they are happening or before they happen with each batch. “We have a data-driven mind-set now,” Penland said. “We are asking, ‘How do we interpret this, and what is the data telling us?’ rather than having a more reactive knee-jerk response, fighting fires all the time.”
For more information about Boehringer Ingelheim and the PI System, watch the full presentation here.