基于光谱和色谱数据融合策略的青叶胆及近似种的鉴别研究

Study on Differentiation of Swertia leducii and Its Closely Relative Species Based on Data Fusion of Spectra and Chromatography

青叶胆(Swertia leducii)为獐牙菜属(Swertia)一年生草本植物,在肝炎病治疗方面效果显著。其与同属近似种外观极其相似,加之常以干燥全草入药,仅从形态难以正确鉴别。不同物种有效成分存在明显差异,其药效也有所不同。基于光谱和色谱数据融合建立青叶胆及近似种的鉴别方法,为青叶胆药用真实性与安全性提供科学依据。采集青叶胆及其近似种植物共102份样品的傅里叶变换红外光谱(FTIR)和超高效液相色谱(UPLC)指纹图谱;利用标准正态变量(SNV)、多元散射校正(MSC)、 Savitzky-Golay平滑(SG)、一阶导数(1D)、二阶导数(2D)等方法对原始红外光谱数据进行预处理,通过系统聚类分析(HCA)探讨獐牙菜属不同种类样品化学信息相似性与差异性;Kennard-Stone算法将所有样品按2∶1比例划分为训练集和预测集,训练集基于FTIR, UPLC,低级与中级数据融合建立随机森林(RF)判别模型,预测集用于验证模型预测能力,其中灵敏性(sensitivity)、特异性(specificity)、精密度(precision)和正确率(accuracy)用来评价模型性能。结果显示:(1)采用SNV+SG+2D组合对FTIR数据进行预处理,R^2Y和Q^2最大,分别为91.2%和84.1%,所有类别被正确区分,为最佳预处理。(2)HCA反映了5种獐牙菜属植物样品分类情况与亲缘关系,除紫红獐牙菜外,其余4种獐牙菜植物均分类正确,准确率为93.1%;青叶胆、川东獐牙菜、紫红獐牙菜与西南獐牙菜亲缘关系较近。(3)基于FTIR、 UPLC、低级和中级数据融合策略建立RF判别模型,样品错判总数分别为1, 5, 1和0,中级数据融合效果最佳,所有样品均正确分类,所建模型性能良好。FTIR与UPLC通过中级数据融合策略结合RF判别分析能正确鉴别不同种类獐牙菜属植物,结合HCA分析能够明确青叶胆及其近似种之间的亲缘关系,为獐牙菜属植物资源开发与质量控制提供理论基础。

Swertia leducii is an annual herbaceous plant of the genus Swertia. It has a remarkable high effective in treating liver inflammation. The appearance of S. leducii and the species of the same genus is very similar, and the whole dry herb of Swertia plants is often used as medicine. It is very difficult to correctly identify different species from the morphology. Nevertheless, It is different in treating effective due to different species with different chemical components. In this study, based on data fusion of spectra and chromatography, an effective identification method of S. leducii and its closest relative species was established to provide the scientific basis for authenticity and security of S. leducii medication. Fourier transform infrared(FTIR) and ultra performance liquid chromatography(UPLC) of 102 samples of Swertia were collected from 5 species. Standard normal variate(SNV), multiplicative signal correction(MSC), Savitzky-Golay smoothing(SG), first derivative(1 D) and second derivative(2 D) were used to treat raw spectral data. Then, the optimal spectral data was utilized to process Hierarchical cluster analysis(HCA) for analyzing the similarity and dissimilarity of genus Swertia with different species. Kennard-Stone algorithm was applied to divide 102 samples into the calibration set and validation set in accordance with 2∶1 ratio. The calibration set was established the random forest(RF) discriminant model basing on FTIR, UPLC, low-level and mid-level data fusion, and the validation set was used to test the predictive ability of these models. In addition, the model performance was evaluated by sensitivity, specificity, precision and accuracy. The results indicated that:(1) SNV+SG+2 D was the optimal pretreatment that all samples were correctly classified with the highest R^2Y(91.2%) and Q^2(84.1%).(2) HCA could reflect the classification and genetic relationship of S. leducii and its wild relatives. The other 4 species excepting S. punicea were correctly classified and its total accuracy rate reached 93.1%. S. punicea, S. cincta and S. davidii had closed relationship with S. leducii while S. angustifolia was relative far.(3) Comparing the FTIR, UPLC, low-level data fusion and mid-level data fusion, the number of error samples in the classification of RF analysis were 1, 5, 1 and 0, respectively. In the RF models, the best classification of mid-level data fusion with none error samples was better than other data matrices. Mid-level data fusion combined with RF methods can identify different species of genus Swertia and display the genetic relationship between S. leducii and its wild relatives. Besides, it could provide a theoretical basis for the development of plant resources and quality control of genus Swertia.