Espectrocospia FTIR, HPLC e Redes Neurais Artificiais para determinação analítica da lactose residual e outros açucares no leite

Abstract

Lactose is the sugar naturally present in milk and has great nutritional importance, but this saccharide is not well absorbed by most of the world population, due to the decrease in the production of the lactase enzyme, which acts in the small intestine, hydrolyzing the lactose ingested in two monosaccharides, glucose and galactose. The enzymatic hydrolysis of lactose by industries offers a variety of products with low lactose. The methodology used to measure the concentration of sugars in milk after hydrolysis is preferably liquid chromatography (HPLC) with refractive index detection (RID), due to its ability to distinguish sugars. Mid-infrared spectroscopy lacks this distinction due to the absorption of light by sugars occurring in the same region of the spectrum. In this work, the association of the FTIR (Fourier Transform Infrared) spectroscopy with the tools of artificial intelligence for deep learning was proposed to identify and quantify residual lactose and other sugars in milk. Convolutional neural networks were built from the mid-infrared spectra, without preprocessing the data, to solve problems in the classification of milk samples, and to predict which sugar class was present, and to predict the sugar concentration values in milk, a regression model was proposed. Raw, pasteurized and UHT milk were added with lactose, glucose and galactose at six concentrations (0.1, 0.5, 1.0, 3.0, 5.0 and 7.0%) and, in total, 342 samples were submitted to artificial neural networks to identify and quantify the sugars in milk. In addition, the enzymatic hydrolysis of lactose by β-galactosidase was performed on 20 samples of pasteurized milk in four different treatments (control and three levels of hydrolysis) which resulted in the simultaneous presence of the three sugars (lactose, glucose and galactose) for quantification by HPLC-RID and comparison with results predicted by convolutional neural networks. The result of the challenge of these samples to classification and regression algorithms indicated a predictive capacity of 80% for classification and 75% for quantification of sugars in hydrolyzed samples. The association of FTIR spectra to artificial intelligence tools proved to be a promising methodology for milk quality control.