【作者】 白立飞；

【导师】 王洪艳；

【作者基本信息】 吉林大学 ， 分析化学， 2006， 博士

【摘要】 将化学计量学方法用于中草药多组分测定和真伪鉴别的研究中,改进的人工神经网络方法(ANN)为中草药多组分同时测定提供了方便,模式识别则实现了中草药的真伪鉴别及质量评价。提出了一种双网络ANN算法,用其处理中草药紫外吸收光谱,提高了网络的自学习,自判别和自拟合能力。为了解决中草药复杂体系背景干扰的问题,首次提出虚拟组分校正的方法。将所有干扰组分归为一个组分(其含量是可调的,称其为虚拟组分),通过ANN的双网络对虚拟组分自修正、自拟合,消除了干扰组分的影响,实现了中草药秦皮中秦皮甲素和秦皮乙素不经分离的同时快速测定。建立了包括秦皮、紫草、黄芩、大黄、槐花等9种中草药和三黄片等三种复方制剂的色谱指纹图谱数据库,此数据库管理系统可以实现两个功能,一个是药材库的维护;另一个是药材质量的模式识别。在中药指纹图谱评价方法研究方面,通过对欧式距离、夹角余弦、峰匹配度等相似性测度的比较,提出了总相似度的概念,结果证明总相似度对样品的分类更加符合实际情况。提出对保留时间的校正和重叠度的计算,减小了保留时间漂移对识别结果的影响,使识别中色谱峰的匹配表征更准确,提高了中草药识别的准确性。论文还提出用共有峰总面积比(待识别样品的共有峰总面积/标准样品的共有峰总面积)来评价中药的整体质量,在识别的基础上实现了黄芩,秦皮等中草药的质量评价,此方法可推广到任何一种中药的识别中。更多还原

【Abstract】 Chemometrics is an interdisciplinary subject which applies statistics,mathematics and mathematical statistics to chemistry. With the development ofcomputer science, more and more people have been attaching importance tochemometrics in recent years. In this paper, chemometrics method has been usedfor the determination of multi-components in Chinese herbal medicine and thepattern recognition of chromatographic fingerprints. The modified artificial neuralnetwork(ANN) method provided convenience for the simultaneous computationof multi-components, and the pattern recognition made the recognition and thequality evaluation of Chinese herbal medicine come true.1. Artificial neural network applied to simultaneous determinations ofmulti-components in Chinese herbal medicineBased on the traditional BP network, a simulation network was added. So adouble ANN(DANN) including training and simulation network was established.Suitable training sample sets could be dynamically chosen by the back simulationof the input absorbency with the simulation network. The capability ofself-studying and self-recognition were improved in this way, and the predictedaccuracy of multicomponent’s concentration was improved greatly in thecomplicated Chinese herbal medicine system.The effect of interferential components was eliminated with the addition ofthe virtual component. So the simultaneous determinations of multi-componentsin Chinese herbal medicine were performed by artificial neural network-UVspectrometry. The virtual component means that immeasureable unknowncomponents are transformed into one or several virtual components, then they aregiven certain concentration values in a certain range and added into studyingsample to join the training and studying of training network, and they were testedusing the simulation error between the simulative absorbency value and theexperimental absorbency value in testing module to judge the rationality of virtualcomponent.In the prediction of actual samples, the experimental absorbency values ofthe samples were input the network, the training sample set was arbitrarily chosen,and the content of virtual component was adjusted to make the simulation error besatisfied with the request. If the error is still unsatisfied with the reuest and closesto a stable value after several adjustment, another training subset is needed. Thecontents of aesculin and aesculetin, extracted from 21 Cortex Fraxinis, werepredicted. Compared the results with those of HPLC, the predicted accuracy wasmore than 90% within the relative errors less than 10%. The measurementprecision of the aesculin and aesculetin in this method were 0.37% and 1.5%respectively.In order to inspect the superiority of DANN, we compared it with PLS. Atraining set which is the same as the one of DANN arithmetic was as the standardmodel of PLS to predict the concentrations of 21 samples. For aesculin, the linearcorrelation coefficients between the predictive contents of ANN and PLS and thedetermined content of HPLC were 0.9914 and 0.8324, respectively. For aesculetin,the linear correlation coefficients were 0.9920 and 0.9204, respectively. Comparedwith the measurement result of HPLC, the predictive accuracy of PLS was 76%within the relative errors less than 10%. The practice showed that the modifiedDANN arithmetic is superior to the PLS arithmetic.2. Establishment of chromatographic fingerprints database of TraditionalChinese MedicinesChromatographic fingerprints database including 9 kinds of Chinese herbalmedicine such as Scutellaaria baicalensis Georgi, Cortex Fraxini etc and 3 kindsof complex prescription such as San Huang pian etc was established. Thisdatabase has two functions: One of the two functions maintenance of the database.We can search basic attribute of Chinese herbal medicine with different names,such as producing area, chemical component, the properties and flavors, specificselection of channles, the illustration of crude drug, analytical method of crudedrug, attending function and so on, it can also finish the addition, revision anddeletion of the medicinal information;the other function is quality patternrecognition of Chinese herbal medicine, the software about the pattern recognitionsystem of chromatographic fingerprint of Traditional Chinese Medicines wasdeveloped using computer technology, and the identification and qualityevaluation of Chinese herbal medicine were realized.3. Chemistry pattern recognition of chromatographic fingerprints of Chineseherbal medicineChromatographic fingerprints of Chinese herbal medicine about Scutellaariabaicalensis Georgi and Cortex Fraxini etc was established, and the automaticchemistry pattern recognition of the fingerprints was achieved. In this method, ifthe integral report of two Chromatographic Fingerprints is provided, the peakswill be automatically matched, the common peaks and unique peak will beautomatically judged, and the corresponding value will be filled in thecorresponding position of the vector to compute comparability. Then theidentification and quality evaluation of drug will be processed.In the studying of evaluation method of the similarity of fingerprints ofChinese herbal medicine, the total similarity ( Q = α ×Cir /dir,α is the matchingdegree of peaks, Cir is the similarity of cosines of the crossing angle , dir isEuclidean distance) was presented based on a large amount of the calculation ofactual samples by the comparison of several similarity measures which includeEuclidean distance, cosine of the crossing angle and the matching degree of peaks.The results also indicated that the total similarity fitted the actual situation betterfor the categorization of samples. And also, the conclusions are as follows:Q>1,the same drug with similar quality;0.7 <Q<1, the same drug;Q<0.7, differentdrugs。In the discrimination of common peaks, the adjustment of retained time andthe calculation of the overlapping degree of corresponding peaks were raised. Sothe effect of the shift of retained time on the recognition was reduced in thechromatographic separation, and the matching symbol of the chromatographicpeak was more accurate in the recognition and the accuracy of the recognition ofChinese herbal medicine was improved.In this paper, the ratio of the total area of common peaks(Ai/As, Ai was totalarea of common peaks of unknown samples, As was total area of common peaksof standard samples) was proposed to evaluate the total quality of Chinese herbalmedicine. Ai/As>0.85, it is excellent, 0.85 > Ai/As > 0.75, it is good, 0.75 >Ai/As >0.60, it is middle, Ai/As< 0.6, it is inferior. Quality evaluation of Scutellaariabaicalensis, Georgi and Cortex Fraxini etc was realized based on the recognition.The recognition accuracy achieves 87.5%. A convenient and feasible method wasprovided for the quality evaluation of Chinese herbal medicine, it can be used inthe identification of any kind of Chinese herbal medicine.更多还原