Compositional analysis of alternative protein blends using near and mid-infrared spectroscopy coupled with conventional and machine learning algorithms

The non-invasive real-time analysis of the composition of alternative, plant-based protein sources is important to control high moisture extrusion processes and ensure the quality and texture of the final extrudates used in the elaboration of meat analogues. This study aims to analyse the composition and presence of gluten in blended plant-based alternative protein sources from pulse, cereal and pseudocereal origin by means of near infrared spectroscopy (NIRS) and mid infrared spectroscopy (MIRS) using conventional and machine learning algorithms. Blends were prepared using five alternative protein sources (barley, wheat, fava bean, lupin, and buckwheat) and spectra were acquired using a low-cost and a benchtop near-infrared spectrometer, and a mid-infrared spectrometer. Using the acquired spectra, partial least square regression (PLSR), support vector machine discriminant analysis (SVM-DA), partial least square discriminant analysis (PLS-DA), and convolutional neural networks (CNN) were used to develop predictive models to determine the composition and to identify samples containing gluten. The protein, moisture, carbohydrates and fat content in blends of alternative protein sources was determined with a RMSEP of 1.59, 0.18, 1.41, and 0.19 %, respectively, when using the benchtop NIR spectrometer and PLSR. Gluten-free samples were identified with high sensitivity (0.85) and accuracy (0.93) using PLS-DA. The study demonstrated that infrared spectroscopy can be used to analyse the composition of blends of alternative protein sources including pulses, cereals, and pseudocereals, as well as to identify gluten-free samples.