For a long time, Takeda has worked closely with Minitab to address manufacturing problems, recognize and act on opportunities for improvement, and conduct effective experiments to achieve process enhancements.

Takeda Pharmaceutical Company Ltd., is a global biopharmaceutical company headquartered in Tokyo dedicated to enhancing the health and well-being of people worldwide,

The Challenge

In Switzerland, Takeda Neuchâtel produces three drugs that treat two types of blood clotting disorders or deficiencies. The medicines are known as “recombinant treatment”; this implies they are manufactured without any human or animal additives. They are made using biotechnological processes that involve cell culture, resulting in pharmaceutical products that offer safety, purity and effectiveness.

Takeda aimed to boost the output of these drugs by improving the efficiency of their production process. To achieve this goal, the researchers needed to investigate which process parameters influenced the performance of the cell culture.

The Solution from Minitab

Takeda’s team gathered data on 30 process parameters that could impact the blood clotting protein during cell culture. They employed various statistical techniques, including feature engineering methods, to prepare the data for analysis. The team discovered that Partial Least Squares was effective in identifying the crucial factors that impact the yield. Yield is a regression technique that models the relationship between multiple predictors and one or more continuous responses. It is especially helpful when predictors are highly correlated or when there are more predictors than observations. The standardized coefficients visualization showed a few significant process parameters that the team identified using this method.

However, Takeda’s team recognized that sharing these results could be challenging because Partial Least Squares regression is relatively advanced. Takeda scientists and engineers are trained and enabled to analyze data independently using Minitab Statistical Software. The team decided to assess how a CART decision tree could supplement the Partial Least Squares regression analysis. The advantages of this approach were: CART trees could validate the PLS results and are easier to use and understand.

A beneficial output of a CART tree is a Relative Variable Importance plot, which ranks the variables according to their significance. The results confirmed the Partial Least Squares approach and sparked fascinating discussions about the yield process. CART also provides a single decision tree, which is another straightforward visualization.

The Results

Researchers utilized multiple machine learning models to gain a better understanding of the yield process and identify critical parameters. The results obtained from the CART decision tree approach were consistent with those from the Partial Least Squares regression approach. Due to their ease of use, CART decision trees can speed up the understanding of the results at Takeda. CART decision trees are intuitive and can be an additional tool for those already familiar with statistical tools.

Takeda’s team plan to enhance the model’s accuracy by integrating advanced decision trees, using Minitab Statistical Software.

We are Minitab’s Value Add Resellers, if you need a quote contact us via email minitab.sales@bowriversolutions.com or visit www.bowriversolutions.com for more information.