How Predictive Modeling Can Benefit Bioreactor Performance

As the biopharmaceutical industry moves towards automated processes, technologies that allow biomanufacturers to predict outcomes and troubleshoot issues is becoming increasingly useful. Predictive modeling is one such technology that may be used to optimize bioreactor performance.

The issue of bioreactor control has become increasingly important because of the difficulty inherent in controlling the behavior of the living cells used in cell culture systems (1). Despite having a well-designed control scheme, a bioreactor can still exhibit poor performance. As such, an approach that includes predictive modeling can be an advantage in developing process controls for bioreactor performance.

Predictive modeling approach

One advantage of using predictive models is that these models enable a data-driven approach, which can assist in accelerating process development, says Tania Pereira Chilima, chief technology officer, Univercells Technologies. Chilima explains that these models allow users to gain a deeper process understanding, which, over time, helps to limit the development resources needed.

“This approach can be used in both batch and continuous processing and can perhaps be even more useful in continuous processing due to the increased number of variables to understand and control (e.g., perfusion rates). Gaining all of these data in real time allows for a faster learning curve and helps to ensure a reproducible process,” Chilima states.

Edita Botonjic-Sehic, director of Analytics at Pall, notes that predictive modeling combines tools and technology “to collect real-time process data and turn it into meaningful information for the long run”. Predictive modeling technology does this by using mathematical algorithms to turn data streams that are pulled from batch or continuous processes into meaningful and targeted assumptions.

“You can think of it like the Amazon algorithm that uses search and purchase data to make certain assumptions and suggestions. Over time, that information becomes richer,” Botonjic-Sehic explains. “When we take this approach to bioprocessing, it allows the prediction of process conditions, such as outcome of glucose feed rate during the process.”

While predictive modeling can be used in both batch and continuous processes, it is, however, currently more challenging to successfully apply to continuous processing, Botonjic-Sehic says. Companies are working to fill in the gaps with batch and continuous predictive modeling solutions that help achieve better outcomes.

Upstream challenges for modeling to solve

The upstream processing stage can be challenging. For one thing, upstream proteins are not pure and need to be differentiated, and on top of that, there are multiple attributes and parameters that need to be monitored and controlled with analytical devices to maintain process conditions, Botonjic-Sehic points out. These parameters include temperature, pressure, dissolved oxygen, pH, and critical quality attributes (CQAs)—all of which need to be part of the model.

“Multi-variant statistical tools are required to come up with the right models because of the complexity and interactions of attributes in the bioreactor; the need for the right tools to measure attributes in real time is a need, and, even when data are acquired, there is still a challenge to translate such data into meaningful information that is ultimately used for prediction,” Botonjic-Sehic states.

Analytical tools such as Raman spectroscopy are commonly used along with other complex spectroscopic techniques to allow for measuring analytes and for predicting information by using pre-determined models. Botonjic-Sehic says that the industry is making the effort to find new solutions to support better predictive modeling.

“The ability to monitor all the CQAs of cell growth using predictive models would be valuable not only in streamlining the use of glucose, nutrients, and metabolites, but also in drastically reducing overall process waste and development time. The ability to predict cell viability and titer concentration is also helpful to aim for the right amount of protein to get pushed [into downstream processing], which is a much purer product,” Botonjic-Sehic explains.

Furthermore, many advancements have been made in the bioreactor space, and Pall itself is also working on its own solutions. “The key here is modeling bioreactors from small scale forward. The ability to demonstrate feasibility early in processes will be critical to lowering costs and increasing accessibility. We need cross functional solutions that enable the measuring, monitoring, predicting, and controlling of processes,” Botonjic-Sehic says.

Design innovation

Predictive modeling approaches in the upstream may also impact design
innovations or accessory development moving forward.

Chilima notes that the industry has definitely seen an increased amount of process analytical technologies (PAT) being developed, for example. PAT enables automated-real time data collection. “Some of these tools also enable different types of data to be collected with a single probe, which is a real advantage. Lately we have also seen significant advances in software development with a number of tools that centralize data collection and analysis to facilitate the development of predictive models, and even digital twins,” Chilima says.

“We expect to see different types of models and approaches that take into consideration a multitude of processes,” adds Botonjic-Sehic. She notes that more work is being done in the space to offer an enabling and standardized platform approach to predictive modeling from the discovery phase through to commercial scale.

Another innovative tool that can enable predictive approaches in upstream processing is the microbioreactor. This is an area on which Univercells has been placing significant focus because microbioreactors can be used to collect an immense amount of data without significant resources, says Chilima. “When the operation of these bioreactors is automated, the costs of developing reliable predictive models can decrease significantly. Taking down the barrier of cost will facilitate more innovation in the space and continue evolving the modality spread in the industry. It is an exciting time to be in development and manufacturing,” Chilima remarks.

“This is a sweet spot for Pall,” emphasizes Botonjic-Sehic. The company continues to build out both its small- and large-scale offerings. “We recognize the industry need for flexible models that streamline development and enable efficient and reliable tech transfer. The idea is to align this flexibility with strong modeling and predictive maintenance functionality from the microbioreactor forward.”

Reference

1. S. Ramaswamy, T.J. Cutright, and H.K. Qammar, Process Biochemistry 40 (8) 2763–2770 (2005).

About the author

Feliza Mirasol is the science editor for BioPharm International.

Article details

BioPharm International
Vol. 35, No. 9
September 2022
Pages: 23–24

Citation

When referring to this article, please cite it as F. Mirasol, “How Predictive Modeling Can Benefit Bioreactor Performance,” BioPharm International 35 (9) 23–24 (2022).