Determination of protein and fat content in fermentation raw materials with NIR reflectance spectroscopy

NIR spectra of sixteen lots of pharmaceutical-grade defatted soybean flour with tightly controlled granulometry were collected and their protein and fat content analysed by standard reference methods. Partial least squares calibration models were developed for both parameters. A new strategy for the use of genetic algorithms for variable selection, weighted genetic algorithm, was tested. This strategy enabled improvements on the predictive ability of the calibration model around 50 %. The spectroscopic method developed will enable faster assessment of raw material quality and will be used to speed up the process of acceptance of new lots.