Detection of Process Fingerprints Applying Multivariate Analysis Methods to Monoclonal Antibody Cell Culture Data

This work presents a sequential data analysis path, which was successfully applied to identify process characteristics in fed-batch cultivation data. The analyzed data set incorporates 116 batch experiments of a fermentation process for the production of a Fc-Fusion protein at 3.5 liter scale. Starting with the univariate analysis of the investigated process and response variables as well as the effect of experimentally manipulated parameters a basic pre-characterization of process is performed.  Consequently, the multivariate analysis mainly based on principal component analysis (PCA) and partial least squares regression (PLSR) is used for further investigation of the dynamic system. Various models are developed and evaluated using different techniques for data-pretreatment, data-recovery, variable transformation, data unfolding and grouping of the experiments. Based on the elaborated models characteristic patterns in the system, process fingerprints, are concluded. Those reveal an understanding for the dynamic importance of the variables for process modeling as well as an understanding for the dynamic predictability pattern of the process based on a tunable incorporation of its history. The obtained results are compared to an analogous analysis for a similar large scale (5000 liter) process data set and multivariate model-based scale-up techniques are discussed. The performance of the linear-based models is compared to alternative statistical techniques such as random forest revealing their opportunities and challenges. Finally, the elaborated fingerprints are interpreted regarding their supportive and simplifying applicability to bioprocess monitoring, analytics and control as well as design of experiments.