Analysis of Total Flavonoid Variation and Other Functional Substances in RILs of Tartary Buckwheat, with Near-Infrared Model Construction for Rapid Non-Destructive Detection

Abstract

According to the requirements of Tartary buckwheat breeding, it is necessary to develop a method for the rapid detection of functional substances in seeds. To ensure a diverse sample pool, we utilized the stable recombinant inbred lines (RILs) of Tartary buckwheat. The coefficients of variation of the total flavonoid, vitamin E (VE), and GABA contents of the RIL population were 15.06, 16.53, and 36.93, respectively. Subsequently, we established prediction models for the functional substance contents in Tartary buckwheat using near-infrared spectroscopy (NIRS) combined with chemometrics. The Kennard–Stone algorithm divided the dataset into training and test sets, employing six different methods for preprocessing spectra. The Competitive Adaptive Reweighted Sampling algorithm extracted the characteristic spectra. The best models for total flavonoid and VE were normalized using the first derivative. The calibration correlation coefficient (Rc) and prediction correlation coefficient (Rp) of the total flavonoid and VE prediction models were greater than 0.94. The optimal GABA prediction model underwent preprocessing via normalization combined with the standard normal variate, and the Rc and Rp values were greater than 0.93. The results demonstrated that the NIRS-based prediction model could satisfy the requirements for the rapid determination of total flavonoids, VE, and GABA in Tartary buckwheat seeds.