ABSTRACT
Biopharmaceuticals commonly require freezing to ensure the stability of the active pharmaceutical ingredients (APIs). At commercial scale, freezing is typically carried out over the course of days in pallets comprising tens of thousands of vials. The selected process conditions have to ensure both complete freezing in all vials and a satisfactory manufacturing throughput. Current process design, however, is mainly experimental, since no mechanistic understanding of pallet freezing and its underlying phenomena has been achieved so far. Within this work, we derive a mechanistic modeling framework and compare the model predictions with engineering run data from the Janssen COVID-19 vaccine. The model qualitatively reproduced all observed trends and reveals that stochastic ice nucleation governs both process duration and batch heterogeneity. Knowledge on the ice nucleation kinetics of the formulation to be frozen thus is required to identify suitable freezing process conditions. The findings of this work pave the way towards a more rational design of pallet freezing, from which a plethora of frozen drug products may benefit. For this reason, we provide open source access to the model in the form of a python package (Deck et al., 2021).