Gluten is a natural complex protein present in a variety of cereal grains, including species of wheat, barley, rye, triticale, and oat cultivars. When someone suffering from celiac disease ingests it, the immune system starts attacking its own tissues. Prevalence studies suggest that approximately 1% of the population may have gluten-related disorders during their lifetime, thus, the scientific community has tried to study different methods to detect this protein. There are multiple commercial quantitative methods for gluten detection, such as enzyme-linked immunosorbent assays (ELISAs), polymerase chain reactions, and advanced proteomic methods. ELISA-based methods are the most widely used; but despite being reliable, they also have certain constraints, such as the long periods they take to detect the protein. This study focuses on developing a novel, rapid, and budget-friendly IoT system using Near-infrared spectroscopy technology, Deep and Machine Learning algorithms to predict the presence or absence of gluten in flour samples. 12,053 samples were collected from 3 different types of flour (rye, corn, and oats) using an IoT prototype portable solution composed of a Raspberry Pi 4 and the DLPNIRNANOEVM infrared sensor. The proposed solution can collect, store, and predict new samples and is connected by using a real-time serverless architecture designed in the Amazon Web services. The results showed that the XGBoost classifier reached an Accuracy of 94.52% and an F2-score of 92.87%, whereas the Deep Neural network had an Accuracy of 91.77% and an F2-score of 96.06%. The findings also showed that it is possible to achieve high-performance results by only using the 1452–1583 nm wavelength range. The IoT prototype portable solution presented in this study not only provides a valuable contribution to the state of the art in the use of the NIRS + Artificial Intelligence in the food industry, but it also represents a first step towards the development of technologies that can improve the quality of life of people with food intolerances.

The most common genotype for beta casein (β-CN) in European dairy cattle is A1A2, producing A1 and A2 variants merged in most traded bovine milks. A premium A2-milk category emerged worldwide through selection of A2A2-genotyped cows to produce added-value milk. Absence of A1 β-CN in bovine milk was assessed by a straightforward, cost-effective bottom-up low-resolution liquid chromatography–mass spectrometry (MS) methodology powered by ion-trap analysis. A preliminary data-dependent MS/MS exploratory approach characterized A1 and A2 variants in milk extracts digested with trypsin, pepsin, and thermolysin. Relative quantitation of selected peptide biomarkers was performed through Selected Reaction Monitoring analysis, detecting levels of 0.1% of A1 β-CN in milk mixtures from A1A1/A2A2-genotyped cows. Besides effective tryptic digestion, alternative thermolysin yielded short peptides that facilitated separation, while acidic pepsin supports the development of low-cost procedures. Affordability and robustness of this strategy favor its implementation by control laboratories addressing the authentication of A2 milk.