Classification of cereal flour by gas ghromatography – mass spectrometry (GC-MS) liposoluble fingerprints and automated machine learning

Abstract

An innovative and rapid approach is described for classifying common types of gluten and non-gluten cereal flour (wheat, rye, triticale, barley, oats, and corn) into the groups defined by their botanical origin. Liposoluble compounds were extracted from flour samples, derivatized, and analyzed using gas chromatography – mass spectrometry (GC-MS). Raw signals used for data processing consisted of mass spectra scans of full chromatograms. These represented unique fingerprints for each class. An automated machine learning framework was applied for classification. The algorithm automatically explored each of the 39 classifiers provided by the software. Using 10-fold cross-validation, a simple logistic classifier was recommended to be optimal. The constructed model resulted in 85.71% correctly classification according to the botanical origin. Furthermore, it unequivocally discriminated samples of non-gluten corn flour. This non-targeted strategy supports the use of artificial intelligence in developing methods for flour authentication.