Determination of protein digestibility in whole sorghum grains by near-infrared spectroscopy

Multiple laboratory protocols have been developed to assay protein digestibility of sorghum [ Sorghum bicolor (L.) Moench] seeds. However, they all involve destructive sampling methods, which render the grains unusable as seed. In this study, near infrared spectroscopy (NIRS) was used to predict protein digestibility in whole sorghum kernels. A non-linear Honigs regression approach was used to develop a calibration model validated in a total of 734 entries. The calibration model, initially developed with 257 samples (training set), had an R 2 = 0.89, root mean square error of cross validation (RMSECV) = 8.69, and standard error of cross-validation (SECV) = 8.71. It predicted the protein digestibility of the test set (n = 29) with R 2 = 0.91, standard error of prediction (SEP) = 8.02 and root mean square error of prediction (RMSEP) = 7.88. The test set was introduced into the model as library samples to form the cross-validated set (n=286), which predicted the external validation set (n =145) with R 2 = 0.67, RMSEP = 13.10 and SEP = 9.35 The model showed strong performance during calibration and cross-validation (RPD > 3) however, external validation resulted in moderate predictive ability (RPD = 1.9) indicating suitability for screening purposes. Validation on independent breeding lines resulted in an RPD of 1.6, further supporting the use of the model for classification and early-stage selection rather than precise quantitative prediction. These results show that the NIR calibration model can reliably be used as a high throughput, non-destructive, pre-screening tool for protein digestibility in sorghum germplasm. • Laboratory protocols for measuring protein digestibility involve destructive sampling. • Near-infrared spectroscopy can be used to predict protein digestibility in whole sorghum grain. • A fast and accurate calibration model was developed for protein digestibility in sorghum. • The Honigs regression model performed well with an R 2 of 0.89.