论文部分内容阅读
目的:基于反向区间偏最小二乘法(BiPLS)和多模型共识偏最小二乘法(cPLS),建立新生儿苯丙酮尿症的FTIR/ATR光谱筛查模型。方法:利用BiPLS将全光谱划分成若干个子区间,优化子区间数目,优选特征子区间组合,然后分别建立基于全光谱和BiPLS优选波段的定量模型,以均方根误差(RMSE)、平均相对误差(MRE)、预测准确率(Acc)、灵敏度(Sens)、特异性(Spec)为指标,比较2种模型的性能。结果:与全光谱相比,BiPLS优选波段将波长压缩了85.0%~90.5%,但所建模型的多项性能指标却有不同程度地提高。在最优条件下,Phe浓度模型的R、RMSE、MRE分别为0.93、87.0、0.27,Sens、Spec、Acc分别为98.2、98.7和98.4;Phe/Tyr比值模型的R、RMSE、MRE分别为0.93、3.55、0.36,Sens、Spec和Acc均为100。Phe浓度模型对接近筛查切值的样本会产生误判,而辅以Phe/Tyr比值模型可将Sens和Spec提高到100。结论:BiPLS极大地压缩了建模所需波段,降低了模型复杂度,提高了模型精度;与cPLS相结合,本文建立的Phe浓度和Phe/Tyr比值的定量校正模型,有望用于新生儿苯丙酮尿症的大人群快速筛查。
OBJECTIVE: To establish a FTIR / ATR screening model for neonatal phenylketonuria based on reverse interval partial least squares (BiPLS) and multi-model consensus partial least squares (cPLS). Methods: By using BiPLS, the whole spectrum was divided into several sub-intervals, the number of sub-intervals was optimized, and the combination of characteristic sub-intervals was optimized. Then the quantitative models based on the full spectrum and BiPLS preferred bands were established respectively. The RMSEs and average relative errors (MRE), prediction accuracy (Acc), sensitivity (Sens), specificity (Spec) as indicators, compare the performance of the two models. Results: Compared with the full spectrum, the preferred band of BiPLS was 85.0% -90.5%, but the performance of the model was improved to some extent. Under optimal conditions, the R, RMSE and MRE of the Phe concentration model were 0.93, 87.0 and 0.27, respectively, and the Sens, Spec and Acc were 98.2, 98.7 and 98.4 respectively; the R, RMSE and MRE of the Phe / Tyr ratio model were 0.93 , 3.55, 0.36, Sens, Spec and Acc are 100 respectively. The Phe concentration model misjudges the samples near the cut-off value, and the Phe / Tyr ratio model increases Sens and Spec to 100. Conclusion: BiPLS greatly reduces the band needed for modeling, reduces the complexity of the model and improves the accuracy of the model. Combined with cPLS, the quantitative correction model of Phe concentration and Phe / Tyr ratio established in this paper is expected to be used in neonatal benzene Rapid screening of large groups of patients with ketorrhea.