Process Variability Source Analysis for a Multi-step Bio-process

Typical manufacturing bio-processes operate as multi-step batch operations. Abnormal variations are often experienced in commercial manufacturing which affects the ability to maintain the product supply to the patients. The low titer is usually highly correlated with high lactate in recombinant Chinese hamster ovary (CHO) cells producing immunoglobulin G, but there are limited means to analyze the vast amount of process data archived from manufacturing sites and understand the causes of the multiple sources of process variability. In this study, we investigate diverse data from a cell culture process, from historical runs of a multi-step manufacturing process for a recombinant antibody. Data includes intra-batch measurements, batch conditions and media preparation measurements. Cell culture process stages from inoculum train to production runs.
To analyze manufacturing variability in multiple stages with various sources of data sets, we first adopt principal component analysis (PCA) based monitoring to find in which space high lactate and low lactate batches can be separated, and apply clustering to identify fault directions for intra-batch data from production runs. We then predict final lactate using data from inoculum stages with a nonlinear support vector machine (SVM) classifier. These models are used to uncover potential causes of high lactate. This data-driven modelling can provide new insights on process characteristics to understand the sources of process variability and thus mitigation procedures can be implemented.