(568ae) Experimental Stress Testing of Multivariate Data Analysis (MVDA) models in Chromatography

As the biotech industry moves toward implementing proactive practices to ensure product quality and process efficiency, improving data analysis is also becoming necessary. Univariate data analysis can often be inefficient and result in misleading conclusions due to the complexity of biopharmaceutical processes. Multivariate data analysis (MVDA) has several benefits over univariate analysis such as increasing sensitivity to process deviations, reducing the time required to analyze large amounts of data and providing more information to more easily review and identify process deviations. These benefits can standardize monitoring practices, increase process knowledge and potentially save batches. MVDA is proving to be a useful predicting tool for different types of processes. However, all of the current applications are tested based on development or manufacturing data without the freedom to stress test the models for potential future deviations. The models are improved only after specific failures or process deviations are observed. In addition, the model capability to detect certain type of deviation is unknown until the deviation happens. Knowledge of model capability increases the possibility to apply MVDA in manufacturing processes. In this work, we looked at downstream purification as a case study and presented the results from stress testing a cation exchange chromatography step and analyzing the process with an MVDA model. Stress testing was performed in the laboratory setting through controlled experiments. The design of experiments included all the typical deviations in chromatographic processes such as load pool and elution buffer variations, column packing conditions and equipment failures. The effects of different modeling configurations to the MVDA fault detection ability were also studied in a systematic fashion. This work highlights the MVDA model capabilities and limitations in chromatographic process monitoring and demonstrates the methodology to evaluate the MVDA model performance for process monitoring and fault detection.