CARACTERIZAÇÃO FÍSICO-QUÍMICA DE RICOTAS VIA ESPECTROSCOPIA NO INFRAVERMELHO E MÉTODOS DE CALIBRAÇÃO MULTIVARIADA

Abstract

Ricotta is a kind of fresh cheese, obtained by the precipitation of the proteins in the cheese whey. According to the current legislation, ricotta is framed on standards of identity and quality of low-fat cheeses, however, studies show a great variability on the centesimal composition. It justifies the necessity of establishing quality standards, and the development of methodologies that allow a fast and efficient control of the product. Besides it, conventional methodologies used to determine the centesimal composition of the ricotta, though they are part of the routine analysis in laboratories of quality control, they are onerous, time consuming and generate residues. In this sense, the objective of this study was to develop a method for the quantification of physical/chemical standards, using Near Infrared Diffuse Spectroscopy (NIRRS), associating to methods of multivariate calibration. For the construction of multivariate models (specially PLSR) it were used the media of concentration of acidity, carbohydrates, ashes, chlorides, fat, pH, protein and moisture, obtained by conventional methodologies (titration method, spectroscopic, muffle carbonization, titration, Gerber, potentiometric, Kjeldahl and gravimetric method, respectively), as well as data of the near infrared spectroscopy. It was collected spectra in duplicate, 33 spectra were used for the calibration phase, and the 5 remaining used to the external validation phase. The best results for fat, protein and moisture levels were obtained on the spectral region between 1100 to 2500 nm. The optimized model for determination of fat used the Multiplicative Scatter Correction (MSC), with 6 latent variable (VLs), acquiring correlation coefficients of Rcal= 0.968 and Rval= 0.936 allowing the quantification of fat with a medium prevision error (Er) of 6.37%. For the protein level, the best result was obtained using MSC and data centered on media (DCM). The model of regression, with 6 VLs, presented correlation coefficients of Rcal= 0.968 and Rval= 0.885, and determination of protein with Er of 5.95%. The best model for determination of moisture used normalization, with 4 VLs correlation coefficients of Rcal= 0.851 and Rval= 0.757 and allowing the quantification of moisture with and Er of 1.91%. It was not possible to build models for acidity, carbohydrates, ashes, chloride and pH parameters, presenting low values of Rcal and Rval, demonstrating the low capacity of forecasting even for samples that compose the calibration set through the proposed methodology. These results demonstrate the potential of multivariate models on determination of fat, protein and moisture levels on samples with complex matrices (ricotta) and also show the advantages of the association NIRRS-PLSR which allows a fast quality control with minimum manipulation of the sample.