Selecting Raman spectra filtering based on an exhaustive statistical approach for inline bioprocesses monitoring using Sf9 insect cells

The spectral preprocessing step of data derived from Raman spectroscopy is important for chemometric models’ calibration, which allows real-time monitoring of pharmaceutical bioprocesses. Thus, the present work aimed to establish, with statistical criteria, the best combination of spectral filters for the biochemical monitoring of the baculovirus/Sf9 insect cell system. The production of rabies virus-like particles was used as a model. Combining moving window smoothing and offset baseline correction spectral filters demonstrated the highest efficacy in biochemically monitoring the baculovirus/Sf9 insect cell system using Raman spectroscopic data and Partial Least Squares regression. However, when applying Artificial Neural Network modeling, a different set of filters proved more effective: wavelet denoise spectral smoothing, asymmetric least square baseline correction, standard normal variate normalization, and quadratic first derivative. The simulation results demonstrate that by following these guidelines for spectral preprocessing, cell viability, glucose, lactate, glutamine, glutamate, and ammonium can be satisfactorily monitored in real-time, but not the density of viable cells. The absolute errors for cell viability, glucose, lactate, glutamine, glutamate, and ammonium were lower than 12%, 0.47 g L− 1, 8.95 mg L− 1, 0.11 g L− 1, 0.10 g L− 1, 3.21 mg L− 1, respectively, which are suitable for bioprocess in-line monitoring and control through a soft sensor.