Multivariate models have been designed in order to be implemented at pilot plant scale. They allow the prediction of the degree of the enzymatic hydrolysis of bovine hemoglobin from on-line mid-infrared (MIR) attenuated total reflection (ATR) measurements. The procedure is very safe and practical, but the spectral information is unfortunately difficult to use. First, remote MIR optical fiber measurements induce a significant energy loss. Secondly, the spectra of proteins in aqueous solutions generally consist of broad and overlapping signals. Principal component analysis (PCA) of the full data set reveals some variations between the different batches as well as definite deviations from the ideal linear situation. This suggests the use of artificial neural networks (ANNs), which have the useful qualities of generalization ability and robustness. Preprocessing of the spectral data (including PCA and wavelet transforms) have been tested. In addition, the reliability of the calibration is ensured by the interpretation of the spectroscopic information. This permits the construction of very parsimonious models and contributes to greater confidence in the analytical tool built for process monitoring. Finally, the estimations of the prediction error are demonstrated to be acceptable. With an unknown proteolysis batch, this error is found to be lower than 0.4 for the degree of hydrolysis covering the range 0 to 8.7.
Cyril Ruckebusch, Bernard Sombret, Rénato Froidevaux, and Jean-Pierre Huvenne, "On-Line Mid-Infrared Spectroscopic Data and Chemometrics for the Monitoring of an Enzymatic Hydrolysis," Appl. Spectrosc. 55, 1610-1617 (2001)