Bio-Pharma

Client:

A US-based biotech/pharma company with operations across the entire North American region with revenues of over $1 billion.

Challenge:

‘Innovate or perish’ seems to be the undercurrent driving most modern business endeavours. Bio-pharmaceutical and biotech companies are no different; they are investing and focusing a great deal on research and product innovation.

Our client’s biggest challenge was to achieve operational excellence in producing live and genetically engineered cells. The primary business challenge was that, despite developing cells under the same parameters and through identical processes, there was a significant variation in the end result of the yield quality. The variation in quality was sometimes up to 60 percent, which led to concerns of quality uncertainty that negatively affected the client operating in a highly regulated industry.

Solution:

MSRcosmos developed a comprehensive analytics solution that helped the customer reduce errors significantly. The solution involved performing statistical, correlative, and predictive analytics on multiple sources of data.

At the time, the cause of concern to business was peculiar in the history of the industry. Therefore, there had to be an end-to-end analysis of the processes as well as the industry itself.

In this preparatory phase, we worked comprehensively on data consolidation, data cleansing, and ensuring data readiness as the single source of truth. Our Big Data CoE (Center of Engineering) team undertook a complete analysis of the challenges faced by the customer, after which it was concluded that a combination of analytics could significantly help in reducing these discrepancies.

Two key sources of data were identified for performing analytics:

• Historical data
• Shop-floor data / Live data

The company had a huge volume of historical data which was typically used for tracking purposes, and not for optimizing operations. Our team applied correlative analytics on top of the historical data to identify relationships among multiple process parameters.

The team applied predictive and statistical analytics, including:

• Moving averages
• Distribution histograms
• Standard deviation
• Clustering analysis

First, we identified the patterns and prioritized data points that had the most predictive power.

Then, artificial neural network analysis was applied to simulate both the structural as well as functional aspects of biological neural networks, which were then used to model complex processes and determine with greater precision how some specific parameters affected productivity. Employing advanced analytics was key to understanding the factors influencing the quality of the end products. Multiple upstream and downstream parameters and their impact on the yield quality were analyzed. Even the raw source of the end product and its impact on quality was analyzed.

Through analysis of live and historical data, we were able to identify eight key factors that were impacting quality leading to variations. This information allowed the customer to stabilize the yield quality by more than 32 percent, while also efficiently working on managing cell storage and media without any additional operational expenditure.

The Results:

• Stabilized quality by more than 32 percent
• Decreased Op-Ex by more than 13 percent
• Increased utilization of data
• Increased business visibility