Quantificação protéica de grãos de feijão por meio da técnica de espectroscopia no infravermelho próximo (FT-NIRS)

Abstract

Beans are an important source of protein. During storage of bean grains changes occur in the integument color and grain texture. In cultivars of the carioca group, the changes in color are more visible, becoming dark due to the oxidation process of proteins present in the integument, which can cause rejection by consumers. The objective of this work was to construct a bean skin protein prediction model using the Near Infrared Spectroscopy (NIRS), with two distinct grain sizes: flour and bran, 10 carioca bean cultivars and 7 black bean cultivars. Total protein quantification was performed by the Bradford method. Prediction analyzes were performed on an FT-NIRS spectrometer. Calibration and validation models were constructed using NIRS spectral data and protein values obtained by Bradford analysis. For the flour and bran samples were used two methods, the cross validation and the test set, in order to compare both models and evaluate the best prediction of protein contents. The models were adjusted by PLS, analyzing the following statistical parameters for cross validation: RMSEE and R² for calibration and RMSECV and R² for validation. For the test set the statistical parameters evaluated were: RMSEE and R² in calibration, RMSEP and R² in validation. Such parameters help in reducing errors, resulting in results with greater reliability. Data regarding total protein concentration in the bean grain integument were submitted to analysis of variance and comparison of means (Scott-Knott <5%) by Genes software. There were significant differences (P <0.05) between carioca and black bean cultivars after storage for 360 days, and the black bean genotypes presented higher values when compared to the carioca group. Both flour and bran presented values of RPD (Standard Error Ratio), in validation and calibration, for cross validation, above 2, being these values acceptable for prediction. The most efficient model for protein prediction in the integument was cross validation, with an R² greater than 0.80, and flour was the most efficient particle size for analysis due to the homogeneity of the samples. There was a reduction in protein content in the first 15 days after grain storage, with an increase up to 60 days and a drastic reduction after 120 days.