(52b) Big Data Analytics in the Advanced Manufacturing of Biopharmaceuticals

This presentation describes the use of process data analytics and machine learning to improve process productivity, reliability, and control in next-generation biopharmaceuticals manufacturing systems. New information streams are described, including the use of hyperspectral imaging to generate gigabytes of real-time data in the form of high order tensors, and the coupling of on-line automated sampling systems with liquid chromatography-mass spectroscopy (LC-MS) to simultaneously measure transient variations in large numbers of distinct small molecules and proteins during manufacturing. Experimental data are presented from multiple laboratories for a variety of bioprocess unit operations, including bioreactors and freeze-drying systems. The use of that data in first-principles, data-driven, and hybrid models are described, with relationships drawn to chemometrics (e.g., partial least squares, principal component analysis), system identification (e.g., canonical variate analysis), pattern recognition (e.g., Fisher and quadratic discriminant analysis), and machine learning (e.g., ensemble learning, elastic net regularization).