Neelanjan Sengupta, PhD, Staff Scientist, Research and Development, Thermo Fisher Scientific
To better control variability and achieve consistency in a specific process, it is important to understand what components in a complete cell culture medium formulation are driving performance. A key driver is a component that has a strong positive or negative influence on process performance and must be within an optimal range for cultures to achieve optimal performance.
Accurately assessing which components in a complex medium are driving a process requires identifying those showing a statistical correlation of performance across multiple lots of a medium, or supplement, of interest. Such analyses are the first step toward identification of key drivers in a bioprocess. Furthermore, parameters that simply correlate with variability need to be differentiated from those that truly cause variability through statistical analysis and experimental data. Finally, for the best process performance, optimal ranges need to be defined for the key drivers.
It is important to remember that each process is different, with its own key drivers, even when similar base media and supplements are used. Using a statistical methodology will correctly identify key components and their optimal ranges for each specific process.
We use a large number of assays to characterize a medium or supplement chemically, identifying and quantifying all its potential components. The ultimate goal is to build a mathematical model to predict which lots will perform well in a given process. The mathematical model uncovers hidden interactions and drivers by including the influence of all components taken together.
With a table of analytical data, we could examine just one component at a time. But by using our unique biostatistical models, which mimic biological behavior, one can consider everything together — reducing a large number of components to just a few key drivers. This ultimately provides a predictive tool for media optimization and screening.
Using a phased approach
To create a statistically sound model, first we work with a customer to accumulate analytical and performance data on at least 5 to 10 lots of cell culture media or supplements.
- First Phase: We begin with a small data set (developed in the preliminary phase) to build competing initial biostatistical models.
- Second Phase: In this critical phase, we determine the causative nature of key drivers. Many components can vary performance, but to control process variability, we need to find the key drivers. The output of the second phase provides a recalibrated model, and we will use it to confirm the key drivers before moving on to the third phase.
- Third Phase: Here, final model verification generates a predictive tool to test each cell culture medium or supplement lot. Using new lots of media or supplements, we perform analytical testing to evaluate the model. Then the customer evaluates those formulations in a small-scale version of the actual process. If experimental data match the prediction, then we have a locked-down model; if we don’t have an exact match, then we may recalibrate the model slightly.
Bioprocess variability can be caused by very small changes in specific components or impurities in chemically defined media or supplements. Always consider all sources when you’re thinking about controlling those variabilities. It’s important to build a statistically significant data set to elucidate those components responsible for observed variability. Through predictive mathematical modeling, we can achieve consistent production performance.
Join Dr. Neelanjan Sengupta for an exclusive November 18 webinar focused on the KDI predictive modeling approach. Register today.