基于化学指标的烟叶产区正交偏最小二乘判别分析

The OPLS-DA Model of Tobacco Producing Areas Based on Chemical Measurements

为构建基于化学指标的不同产区烟叶的模式识别模型,收集2003—2007年5个年度四川、云南和福建3省份共计1262份烤烟样品的21种化学指标,并进行正交偏最小二乘判别分析(OPLS-DA)。结果表明,基于21种化学指标所构建的OPLS-DA模型显著可靠,模型参数R^2X、R^2Y和Q^2值分别为0.429、0.702和0.627;模型提取出2个预测主成分,对四川、云南和福建3省未知烟样识别正确率分别为76.67%、93.33%和93.33%。基于化学指标的烟叶产区OPLS-DA模型,可以有效识别不同产区烟叶样品,筛选出各产区特征指标并解释产区间差异。

In order to construct the pattern recognition model of tobacco producing areas based on chemical measurements,21chemical measurements from1262samples of flue-cured tobacco were collected from2003to2007,and were used to construct aOPLS-DA model.The results showed that the OPLS-DA model was highly significant with the following model performance statistics(R2X=0.429,R2Y=0.702,Q2=0.627).The OPLS-DA of the training set gave a model with two Y-predictive components,and gave thecorrect rate of76.67%,93.33%and93.33%to distinguish samples from Sichuan,Yunnan and Fujian tobacco producing areas.TheOPLS-DA model based on chemical measurements can recognize tobacco samples from different producing areas,and can be used inscreening for biomarkers for different areas and providing interpretations for regional difference of tobacco samples.