This work assessed the impact of laser wavelength and sample conditioning on offline monitoring (viable cell density, cell viability, virus titer, glucose, lactate, glutamine, glutamate, and ammonium) of SARS-CoV-2 virus-like particles production upstream stage by Raman spectroscopy. The evaluated chemometrics techniques were Partial Least Squares (PLS) and Artificial Neural Networks (ANN), and different spectral filtering approaches were also considered. ANN showed better prediction capacity for most of the parameters, but ammonium and lactate, better predicted by PLS, and glutamine, no difference between modeling techniques was detected. For cell growth parameters and virus titer, samples without cells measured at 785 nm Raman laser wavelength originated better-adjusted models. This laser wavelength was also more appropriate for the remaining monitored experimental parameters except for glucose, in which the best model came from the spectral database acquired at 1064 nm wavelength. Cell remotion significantly increased the accuracy of viable cell density, cell viability, glutamate, and virus titer models. However, glucose, lactate, and ammonium models showed better prediction capacity for samples containing cells. Thus, it was demonstrated that laser wavelength, sample conditioning, spectral preprocessing, and chemometric modeling techniques need to be tailored for each experimental parameter to improve accuracy.
Read full abstract