Being able to predict the product yield at all stages in a long-running mammalian cell culture manufacturing process in an industrial context is vital for both operational efficiency and process consistency. We used Raman spectroscopy to quantitatively monitor (in terms of glycoprotein yield prediction) a fed-batch fermentation from start to
finish. Raman data were collected from 12 different time points in a Chinese Hamster Ovary (CHO) based manufacturing process and across 37 separate production runs.
Variable selection methods were used to generate accurate prediction models for final glycoprotein yield (relative error of predictions less than 3%)
for every stage of the bioprocess from 100L, up to the final 5000L bioreactor. The approach of using a
Raman based model to predict potential product yield and quality outcomes could have value and be useful for managing large scale manufacturing campaigns.