The Use of Signal Filtering Algorithms in Bioreactor Characterisation and Monitoring Using Raman Spectroscopy

by Giuseppe Elia, Dylan Jones, Matthew Harding, and Chris Whitmore
Volume 14, Issue 3 (Fall 2015)

Raman spectroscopy offers an attractive solution for monitoring key process parameters and predictive modelling in cell culture processes using transgenic Chinese hamster ovary (CHO) cells. Frequent in-line measurements offer the potential for advanced control strategies. However, an erroneous value created by analytical signal noise is a significant issue that can affect process controls negatively. One such challenge is to differentiate the signal reflecting process changes, ranging from random to gross error, in a timely manner so the process control system can respond to these changes and maintain adequate control. The frequency of measurement acquisition in Raman monitoring makes signal filtering a viable solution to the problem of erroneous measurements. In this study, partial least squares (PLS) models were developed for multiple process parameters (such as glucose, glutamate, and viable cell density) using data from a 10 L bioreactor. The PLS models were applied to over 10,000 spectra taken at approximately five-minute intervals. Signal filters were applied to clean the resulting prediction data. The resulting predictions showed far fewer fluctuations from random errors, as well as greater resistance to gross (malfunction-based) errors, than the non-filtered prediction data. Effective signal filtering could represent a major improvement in the reliability of in-line spectroscopic monitoring of bioreactor processes and greatly improve the potential for robust control strategies on those processes...

Elia G, Jones D, Harding M, Whitmore C. The use of signal filtering algorithms in bioreactor characterisation and monitoring using Raman spectroscopy. BioProcess J, 2015; 14(3): 34–43.

Posted online October 9, 2015.