Application of machine learning in ensuring viral safety of biotherapeutics: Case study demonstrating prediction and optimization of viral clearance performance of anion exchange chromatography

Abstract

This study establishes the applicability of various machine learning (ML) algorithms for creating in silico models based on data from historical bioprocess development and manufacturing activities. We demonstrate this by developing ML-based models for a particular case of predicting viral clearance performance of anion exchange (AEX) chromatography operated in flow-through (FT) mode, a processing step that is essential for ensuring viral safety of biotherapeutics and that requires significant resources due to a large number of expensive wet lab design and characterization experiments. Using these ML-based models, we then evaluate the impact of various feed-, virus-, and process-related parameters (such as the molecular attributes of the product, the type of chromatography matrix, loading capacity, equilibration/wash buffer pH and conductivity, virus type and load, etc.) on the viral clearance performance and hence generate process understanding essential for developing FT-AEX chromatography processes with robust viral clearance performance. Finally, we show that ML-based optimization methods can be used to determine process operating ranges for improved viral clearance performance. These results are significant as they provide approaches and understanding necessary for faster and cheaper development of robust FT-AEX chromatography-based viral clearance processes and generally provide guidance on the implementation of ML-based prediction and optimization approaches in bioprocessing - crucial to reducing costs and time associated with the design and manufacturing of biotherapeutics and making these treatments widely accessible.