One challenge pharmaceutical companies face, is moving promising biotherapeutic products into the clinic as fast as safely possible. Devoting the correct amount of time and resources to early stage products can streamline the late stage and commercial development activities. A cell line change between Phase I/II and Phase III can result in spending significant quality, and regulatory resources to demonstrate comparability between the 1st and 2nd generation products. Early DOE screening studies can select for cell lines that have the most robust manufacturability profile, perhaps negating the need for a cell line switch. As the project progresses, process characterization studies can be very labor intensive and are usually delayed until a product has proven itself worthy of investment through success in early clinical trials. Definitive Screening Designs1  have potential to be a one-stop shop for screening important parameters and optimizing settings with low resource costs. Advanced machine learning methods have revolutionized the analysis of large data sets where the data can be divided into portions for training, testing and validation. During bioprocessing development, the datasets are usually small and until now have not been suitable for portioning the data into different training and validation sets. The fractionally weighted bootstrapping method2 uses the same data set for training and validation.  This auto-validation technique has been employed to successfully de-risk model testing experiments that complete the DOE lifecycle.

1  Jones, Bradley and Nachtsheim, Christopher 2011 Definitive Screening Designs with Added Two-Level Categorical Factors  ASQ.org Vol. 45, No. 2, April 2013

2   Ramsey, Phil and Gotwalt, Chris 2017  Model Validation Strategies for Designed Experiments [Power point Slides] Fall Technical Conference https://falltechnicalconference.org/wp-content/uploads/2018_Presentations/3B-FTC-Ramsey-Final.pdf

Learning Objectives:

1. Identify phase appropriate DOE strategies to minimize development timelines and maximize product life-cycle efficiency

2. Recognize the importance of system variability on the DOE settings and levels explored

3. Understand how to use auto-validation to identify model terms