Jiansheng Lu

Process analytical technology (PAT) is a framework that encourages the implementation of analytical tools for real‐time monitoring of critical batch components during the production of pharmaceuticals. In mammalian cell cultures, Raman spectroscopy as a PAT tool has demonstrated its value for accurate in‐line monitoring of various process parameters as well as quality attributes. However, as process development cultures are transferred to manufacturing scales, the small‐scale models are often adjusted either with new calibration data or with transfer learning algorithms. This model adjustment step makes the overall process transfer to manufacturing less straightforward from an analytical standpoint, requiring additional resources and specific data treatment expertise to fine tune the models. In this work, we compare a specific spectra processing approach against existing approaches reported in the literature for processing Raman spectra in the context of scale‐up challenges. The proposed novelty is a preprocessing strategy which computes spectra normalization based on the 3100–3700 cm ⁻¹ band. Data generated at 3 L scale were used to monitor a 3000 L bioreactor without model recalibration or transfer learning with equivalent accuracy as a same scale 3 L bioreactor monitoring. While broader validation case would be warranted, these results demonstrate a direct scale‐up model implementation in large scale manufacturing under the studied conditions.