水稻蛋白质含量NIR模型适配范围的研究

【目的】比较不同类型样品建立水稻蛋白质近红外模型的效果和适配范围。【方法】通过对178份来自“II-32B/岳早籼6号”的重组自交系和496份水稻品种的近红外反射光谱的比较分析，选择其中59个株系和76份品种作为建模样品，采用偏最小二乘法建立基于品种、重组自交系和混合样品的3个蛋白质含量回归模型。【结果】经模型内部交叉验证和对模型外部重组自交系和品种样品的验证结果的比较分析，发现基于分离群体的模型因蛋白质含量范围较窄，样品来源较单一，适应范围仅局限于本群体内样品蛋白质含量预测，而品种和混合模型对群体和品种样品都表现出良好的适应能力，交叉验证决定系数大于0.90，外部验证决定系数大于0.89, 本试验可为近红外建模的样本集选择提供良好的指导意义。【结论】不同类型样品对建模效果有显著影响，品种模型和混合模型的适配范围显著大于群体模型，研究结果不能支持用背景变异较小的样品建立较高精度回归模型的设想。

Study on Adaptability of NIR Models of Protein Content in Rice

After comparison of NIR spectra of 496 rice varieties and 178 RILs from the cross “II-32B/Yue-Zao-Xian No6”, 76 cultivars and 59 RILs were selected to develop three PLS regression models based on cultivar, RIL and mixture samples, respectively. By comparing the effect of validation and prediction, there was found some limitation of model based on RILs due to the narrow range of distribution of protein content. Nevertheless the models based on the representative cultivars or mixture samples had a high coefficient of determination (r2 >0.90) in validation and low absolute error (<1.1%) in prediction, which should be suitable for primary screening on protein content of rice germplasm and breeding lines.