Modeling The Bioactive Components In Angelica Based On Spectrum Data Fusion

Angelica Sinensis,is an herb used in Chinese medicine to enrich blood,promote blood circulation and modulate the immune system.Due to the variety and complexity of the chemical substances contained in Angelica sinensis and their varying content of components,it is important to carry out accurate and reliable modelling analysis of the main components for the quality control of it.Modern quantitative analysis of traditional Chinese medicine depends on their spectra.As different spectra focus on different chemical substances,it is difficult for one kind of spectrum to fully express the whole information of a sample.Thus,there is a trend to use data fusion techniques among different spectra for quantitative analysis of Chinese medicine.In this thesis,a novel data fusion procedure was proposed to effectively fuse the complex Ultra-Performance Liquid Chromatography and Near Infrared spectra of Angelica samples to establish the quantitative calibration models for the main components of Angelica sinensis.The two spectra were firstly partitioned into block variable subspaces using Fisher’s optimal partitioning method and sure independent screening(SIS)respectively.After this,the two spectra were bottom fused to by putting them together serving as the whole explanatory variable space.Then a consensus model was built on the new explanatory variable space.The consensus model made use of the common information of the fused spectra.To avoid reuse of information in subsequent steps of the procedure,the used common information is deducted.In order to use the specific information to increase the accuracy of the model,recursive boosting iterations are performed between the residual information of each block variable subspace and residuals of response.The final model was obtained by combining the consensus model and the additive boosting model.Finally,the block variables were analyzed according to their importance using stochastic permutation to explain the model.Overall,in this thesis,the quantitative calibration models were built for the content of four major components of Angelica sinensis,including ferulic acid,ligustilide,chlorogenic acid and chuanxiong lactone I.The results show the advantages of the new data fusion procedure proposed over traditional methods.In addition,the obtained block variables helped us to understand the sensitivity and importance of a certain interval of the spectrum to the content of the components of Angelica sinensis,which is important for the interpretation of the model.