Original Discrimination Of Rhizome Gastrodiae Based On HPLC Fingerprint And Near-infrared Spectroscopy Combined With Chemometrics

Gastrodia elata as a well-known traditional Chinese medicine(TCM)product,studies on the quality control of gastrodia elata are increasing in recent years.Due to the complexity of chemical composition in traditional Chinese medicines,how to maximize extraction and separation of the active ingredients is a big hassle that puzzled the researchers.Generally,the quality analysis of gastrodia elata is confined to a single composition like gastrodine or several active constituents in it,thus it is lack of comprehensiveness and integrity.Fingerprint technology,owing to its integrity,comprehensiveness,relevance,hierarchy and fuzziness,can obtain the whole chemical information of traditional Chinese medicine.Fingerprint tools combined with chemometrics for quality control and evaluation of Chinese medicine is neoteric and effective.In allusion to the defects on the assessment of quality of gastrodia elata,the following two investigations have been carried out in the present thesis:(1)In the present study,high performance liquid chromatographic fingerprint incorporated with boosting partial least squares discriminant analysis(BPLS-DA)was employed for distinguishing gastrodia elata samples from six different origins in China(i.e.,Sichuan,Yunnan,Hubei,Anhuibozhou,Anhuidabieshan and Guizhou).Prior to recognition analysis,baseline correction has been carried out by using adaptive iteratively reweighted Penalized Least Squares(airPLS).Then BPLS-DA,compared with PLS-DA,was employed to fulfill the recognition task.The experimental results show that the gastrodia elata quality can be accurately identified via HPLC-DAD coupled with BPLS-DA.In addition,BPLS-DA can provide superior discrimination ability to a single PLS-DA in the origin discrimination of Gastrodia elata.(2)According to the bias/variance decomposition of the ensemble generalization error,the larger diverisity among the submodels produces the more accurate predictions.Therefore,in this thesis,based on classification tree(CT),a new pattern recognition method called PSOBSTCT was developed by invoking boosting and particle swarm optimization(PSO).In PSOBSTCT,CT was used as the basic learning algorithm,and then a set of CT models was produced on the diverse versions of the original training set by using boosting and PSO was employed to search for subsets of the constructed CT models with the largest diversity.Finally,the predicted output of each selected model was combined by weighted majority vote to produce the integrative prediction.Then PSOBSTCT combined with Near-infrared spectroscopy(NIRS)was applied to discriminate gastrodia elata from six different origins.Meanwhile,in order to check the performance of the proposed method,principal component analysis(PCA),boosting classification tree(BSTCT)and traditional CT were utilized to analyze the spectral dataset.The results showed that the origin of gastrodia elata can be discriminated quickly and accurately by using NIRS incorporated with PSOBSTCT.In addition,the invoking of boosting significantly improves the performance of an single CT.The introduction of PSO to select submoldes with large diversity can further improve the model performance.