Abstract
For implementing Raman spectroscopy as an analytical method in downstream processing, extracting molecular information related to biopharmaceuticals is still challenging due to spectral variations caused by spectrometer, setup and fluorescence. This study explores the potential of the Butterworth filter as a preprocessing method for baseline correction and noise reduction in Raman spectra. We first investigate the Butterworth highpass filter's working principle and its optimization by introducing disturbances to spectral baselines and assessing the cutoff frequency ωc's effect on minimizing baseline variations and enhancing the linear correlation (r2) between Raman signals and protein concentrations. The optimal ωc range (0.004 to 0.008 cm) yields an r2≥0.85, outperforming the Savitzky-Golay derivative filter's 0.68. Further, we explore a Butterworth bandpass filter, adjusting low and high cutoff frequencies, showing an 11.6-15 % improvement in r2 over the highpass design. Our results suggest the necessity of specific cutoff frequency selection when applying the bandpass design to the Raman spectra of individual protein molecules and the method for this selection is discussed. By applying the optimization outputs, we developed chemometric models linking Critical Quality Attributes to the Raman data preprocessed by the Butterworth bandpass filter, covering concentrations up to 25.6 mg/mL for a biopharmaceutical immunoglobulin G (IgG) antibody and 4.2 mg/mL for Transferrin. When validated in Cation Exchange Chromatography runs with gradient lengths of 5 and 10 column volume for in-line predictions, the models show high predictability, achieving a coefficient of determination R2 of 0.99 for IgG and 0.95 for Transferrin.
