Comparison of Near Infrared Spectroscopy Models for Determining Protein and Amylose Contents Between Calibration Samples of Recombinant Inbred Lines and Conventional Varieties of Rice

The near infrared spectra of 178 recombinant inbred lines （RILs） from the cross of Ⅱ-32B/Yuezaoxian 6 （YZX6） and 511 varieties in rice were acquired. A total of 80 RILs and 96 cultivars were selected as modeling samples by comparing the spectra similarity primarily. Three partial least square （PLS） regression models were developed, based on the RILs （RIL-model）, the varieties （Var-model） and their mixture （Mix-model）, for protein content （PC） and amylose content （AC）, respectively. Cross validation and outer prediction showed that the models were largely influenced by the range and distribution of modeling samples. The regression model of PC based on the cultivars and the model of AC based on RILs had higher coefficient of determination （r^2 ≥ 0.9） and lower root mean square error of cross validation （RMSECVs）. The disadvantages of RIL samples for PC model and variety samples for AC model were probably caused by the narrow range of variance. Aberrant predictions were obtained for outer sample with PC or AC outside the range or within the distribution gap of modeling samples. The Mix-models gave more reliable prediction as the distribution of RIL and variety modeling samples were complementary to each other.

Comparison of Near Infrared Spectroscopy Models for Determining Protein and Amylose Contents Between Calibration Samples of Recombinant Inbred Lines and Conventional Varieties of Rice

The near infrared spectra of 178 recombinant inbred lines (RILs) from the cross of Ⅱ-32B/Yuezaoxian 6 (YZX6) and 511 varieties in rice were acquired. A total of 80 RILs and 96 cultivars were selected as modeling samples by comparing the spectra similarity primarily. Three partial least square (PLS) regression models were developed, based on the RILs (RIL-model), the varieties (Var-model) and their mixture (Mix-model), for protein content (PC) and amylose content (AC),respectively. Cross validation and outer prediction showed that the models were largely influenced by the range and distribution of modeling samples. The regression model of PC based on the cultivars and the model of AC based on RILs had higher coefficient of determination (r2 ≥ 0.9) and lower root mean square error of cross validation (RMSECVs). The disadvantages of RIL samples for PC model and variety samples for AC model were probably caused by the narrow range of variance. Aberrant predictions were obtained for outer sample with PC or AC outside the range or within the distribution gap of modeling samples. The Mix-models gave more reliable prediction as the distribution of RIL and variety modeling samples were complementary to each other.