Data Mining in Vaccine Manufacturing: Finding Needles in Biological Haystacks

• Activity Number: 365
• Type: Roundtables
• Date/Time: Tuesday, August 2, 2011 : 12:30 PM to 1:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Nelson Lee Afanador*+
• Companies: Merck Sharp and Dohme Corp.
• Keywords: Data Mining ; Random Forest ; Recursive Partitioning ; Vaccine Manufacturing

Abstract

Vaccine manufacturing is the latest field in which advanced data mining methods are beginning to gain widespread acceptance among data analysts. Vaccine manufacturing itself is a complex biological process composed of hundreds of steps carried out over several months. During the manufacture of a single vaccine lot hundreds of processing variables and raw materials are monitored. The challenging aspects with respect to mining biological manufacturing process data begins with obtaining a suitable dataframe which to analyze, proceeds to inherent variation in raw material composition and processing conditions, and ends with high inherent variation in the measurement systems. As such, identifying the root cause candidates for changes in cellular growth or viral propagation is extremely challenging. Given the large numbers of available candidate variables the traditional methods of univariate statistical process control charting, analysis of variance, and least squares regression leave many questions unanswered. Methods such as Random Forest and single-tree recursive partitioning have proven to be important methods in helping drive at potential root causes for observed changes.