Investigations in Raman Spectroscopy as a PAT tool applied in cultivations with Pichia pastoris

Abstract

The present work investigated the applicability of in-line and off-line Raman spectroscopy for the quantification of glycerol, methanol, cell density, and total protein in fed-batch cultivations of Pichia pastoris. For assessment of the multivariate Raman spectra, data pre-processing and orthogonal projections to latent structures were applied. In the context of this work, the optimal pre-processing tools were baseline correction by 1st derivation and linear correction, scatter correction by standard normal variate, and noise removal by Savitzky-Golay filter with 9 points in each moving polynomial. The models yielded root mean square errors of cross-validation of 13.7 %, 13.3 %, 6.24 %, and 11.5 % for glycerol, methanol, cell density, and total protein, respectively. The lowest cross-validation errors for in-line as well as off-line were achieved with prediction of cell concentration. In contrast, the highest error was obtained by glycerol prediction with 18.8 % by in-line Raman spectroscopy. Overall, the in-line Raman probe demonstrated more lower cross-validation errors than off-line Raman spectroscopy. Calibration models were successfully developed for all analytes.