Desenvolvimento de modelos de calibração multivariada em espectroscopia de infravermelho próximo para ácidos graxos em amostras de carne bovina
Abstract
The Brazilian cattle have shown promise prominence in the international market
compared to other commodities. It already knows the important role of fatty acids in the complimentarily of human nutrition. On the other hand, studies have linked the consumption of beef with increasing rate of diseases associated with high levels of "bad" cholesterol (LDL) in the blood. Because of these two aspects on the topic meat (nutritional value and human health), it is seen necessary to determine the levels
oftrans fatty acids in meat. Among the methods for analysis of total lipids and lipid profile, stands out the extraction of fat and chromatography, respectively. In this project was proposed to develop multivariate calibration models for the analysis of total lipids and lipid profile with the use of near-infrared spectroscopy in beef samples
which the levels were previously determined. Eighteen properties were analyzed, including: total lipids and myristic fatty acids, pentadecylic, palmitic, margaric, stearic, palmitoleic, oleic, elaidic, linoleic, α-linolenic and families of fatty acids with branched, saturated, monounsaturated (cis), conjugated linoleic (cis, trans), and omega 3 and 6 polyunsaturated. We used 127 bovine meat samples from the calibration steps and
internal validation (2/3 to 1/3 calibration and validation for internal and 32 external validation samples. We used the PLSR method for the construction of multivariate models. Flesh spectra were extended from 1111 to 1937, and from 2016 to 2500 nm regions, and regions concerning water and noise judged as interfering in the
calibration process were removed during the models construction. We evaluated the
performance of the models based on the applied pre-treatments (smoothing, first derivative Savitzky-Golay, SNV), the number of latent variables, consistency, SEC, SEP and determination coefficients (R2 cal/val). The models chosen as having better predictive capacity were those of total lipids, myristic acid, palmitic acid, margárico and saturated fatty acids, demonstrating that the NIRS has a high potential for the quantification of lipid constituents of the beef.