A Systematic Approach to Process Data Analytics in Pharmaceutical Manufacturing

Abstract This chapter provides a tutorial on the effective application of process data analytics techniques to (bio)pharmaceutical processes. A methodology is proposed in which the process dataset is first interrogated to determine its characteristics in terms of extents of correlation, nonlinearity, and dynamics. The second step is select effective data analytics techniques based on the combination of the identified characteristics. Techniques can be read from a “data analytics triangle,” which has the characteristics at its vertices. This chapter also discusses the value of sparse models and the importance of thorough cross-validation when applying data analytics techniques to (bio)pharmaceutical data sets. Key points and data analytic techniques including lasso and elastic net are described and demonstrated within the context of their application to laboratory- and production-scale data for the manufacture of a monoclonal antibody. A case study illustrates best practices for learning from process data, namely, the importance of maintaining plant-wide connections in the data, implementing procedures to counteract overfitting of the models, and the potential for sparse models in (bio)pharmaceutical operations.