(423b) Applications of Real-Time Machine Learning in Process Performance and Product Quality Prediction in Biopharmaceutical Manufacturing

In recent years, machine learning for predictive analyses has found a foothold in the biopharmaceutical industry. Various algorithms including linear regression, decision tree learning, gradient boosting, and neural networks are evaluated and optimized to deliver predictive capabilities to solve business problems. Models trained on historical data from various data sources (both discrete and continuous) are used to predict response variables of choice. At Amgen, these models are tied into an enterprise data lake in a way that enables predictions to be made in near-real time as data streams from the production line. In combination with domain knowledge, real time machine learning (RTML) has several advantages compared to more traditional prediction techniques and may be suitable for automated root cause analysis as well as driving open- or closed-loop control strategies in a GMP environment. The approach and benefits of RTML will be reviewed through case studies.