NIRS-Based Prediction for Protein, Oil, and Fatty Acids in Soybean (Glycine max (L.) Merrill) Seeds

Abstract

To identify a fast and non-destructive way to determine nutritional traits in soybean, a study was conducted using near-infrared spectroscopy (NIRS) to quantify the oil, protein, and fatty acid contents in soybean seeds. Three hundred soybean accessions obtained from the International Institute of Tropical Agriculture and six varieties were evaluated at two locations in 2021. Fifty random samples of the soybean accessions were scanned over a wavelength of 400–2500 nm at every 0.5 nm interval at the instrumentation laboratory of the Centre for Dryland Agriculture. The spectral data was analyzed using multivariate data analysis software (Unscrambler v9.7). Partial least square analysis was performed on the spectral data and derivative data to determine the best calibration model based on standard error of calibration and R². Goodness of fit was evaluated based on standard error of prediction and the residual percent deviation. Calibration models developed using absorbance gave an R² ranging from 0.991 to 1.000 while that of reflectance ranges from 0.993 to 0.997. Standard error of calibration (SEC) values was between 0.160 and 2.093 for the absorbance groups and 0.166 and 1.376 for the reflectance group. Residual percent deviation (RPD) values greater than 5.0 were obtained using both absorbance and reflectance data for oil and protein, and this signifies that the models were good for quality control and analysis. The result showed an excellent correlation (> 97%) between the predicted and references for all the nutritional traits studied which makes the models good predictors. The developed model was used to predict the oil, protein, and fatty acids of the 306 soybean genotypes, and the observed values were within the reported range for soybean seeds. Thus, NIRS can be used to quantify the nutritional contents of seeds, and it is fast, accurate, and non-destructive.