Discrimination and prediction of polyphenolic compounds and total antioxidant capacity in sorghum grains

Abstract

One hundred and forty-seven sorghum samples, grown in 2015 and 2017, were used to build different near infrared spectroscopic calibration models able to predict total antioxidant capacity, total phenolics, total flavonoids and condensed tannins content. Samples were separated into calibration and validation sets using a nearest neighbours algorithm. The r2pred values ranged from 0.84 (condensed tannins) to 0.95 (total phenolics), whereas the Ratio of Performance to Deviation (RPD) values ranged from 1.9 (total flavonoids) to 3.0 (total phenolics). Comparison of prediction error estimates highlighted the best models with significant differences. Model robustness was tested through a reduction of sample numbers in the calibration set; the highest robustness was found for total antioxidant capacity and total flavonoids. In addition, a partial least squares discriminant analysis model to screen the samples for their tannins level was developed and resulted in good performance; it should be useful to select tannin-free genotypes for the food industry. These models could be used for rapid screening of sorghum breeding genotypes with high antioxidant compounds.