NIR Spectroscopy Model Construction And Application For Amino Acid Composition In Rice Grain

The near-infrared reflectance spectroscopy (NIRS) technique, as a new style physicalmethod, have been developed quickly since 1980s with the advantage of beingnondestructive, fast, economical, environmentally safe analysis and allowing forsimultaneous determination of several traits with one measurement. NIRS has been wildlyapplied in many fields and become the standard methods in some fields or nations. Thepurpose of the present study was to develop and extend the calibration equation for thedetermination of amino acid composition in rice and construct NIRS technique to rapidanalyze the nutrient quality of rice routine analysis and breeding program.A total of 279 representitive samples, selected from original population with 518 brownrice samples, were scanned by using two different adapters with sample weights of about 5 gand 500 mg, respectively. The individual amino acids of brown rice samples weredeterminated by HCl-hydrolyzed HPLC method. Based on the spectra of 3 g brown riceflour, with the pretreatment of 2, 8, 8, 1 and Standard normal variance with de-trending (SNV+D),the NIRS equations for the individual amino acids, except for cysteine, methionine andtyrosine, were well-developed with low standard errors of cross validation (SECV,0.010%-0.063%) and standard errors of prediction (SEP, 0.011%-0.066%), high coefficientof determination of corss validation (1-VR, 87.8%-96.0%) and prediction (R²,83.7%-94.7%), as well as high ratio of standard deviation to standard errors of prediction (SD/SEP, 2.421-4.333). Additionally, calibration equations based on brown rice floursamples were more accurate than those based on brown rice samples.Alkali hydrolysis method was used to determine the content of tryptophan in milledrice due to the destroyed of tryptopan by HCl-hydrolyzed method. The calibration equationof tryptopan, which developed with the pretreatment (1, 4, 4, 1/SNV+D) and the regressionmethod of modified partial least square (MPLS), showed the low standard error ofcalibration (SEC, 0:007%), SECV (0.008%) and SEP (0.007%) and high coefficients ofdetermination RSQ (88.4%), 1-VR (85.0%) and R² (87.1%). The correlations between thereference values and spectra of the sample were studied, and the role played by the NIRabsorbers present in the samples in modeling the calibration equations for tryptophan wasinterpreted by studying the loading of the MPLS factorsA total of 445 samples from two populations (P₁=220 and P₂=225) were used todevelop the calibration equations of the amino acid composition in milled rice. With thepretreatment of (1, 8, 8, 1/SNV+D), the MPLS and ariticial neural network (ANN)algorithms were used to applied to develop the NIRS equations, respectively, and the resultsshowed that the MPLS algorithm was better than the ANN algorithm in this research. TheNIRS equation development were also studied, and the results indicated that the equationbased on two populations is remarkable better than the each equation based singlepopulation, but the calibration, based the selected 189 samples from two populations bySELECT algorithm (NH=0.26), was more similar to the former.By using the NIRS technique to screen the "9311" mutation population, fourty-onelines with higher content of three limited amino acids were primarily identified. Briefly, itwas concluded that the developed equations could be directly used to estimate the aminoacid composition in rice sample, and the NIRS has become the perfect methods for breedingmaterials selection, germplasm evaluation and quality control during brown rice processing.