| As one of the characteristics of traditional Chinese medicine (TCM), Chinese herbal medicine property (CHMP) is the core of traditional Chinese medicine (TCM). The hottest study on CHMP is the correlation between cold/hot property and material composition within a Chinese herbal medicine (CHM) recently.’973program’-’Research of basic theories of CHMP’set up a hypothesis called ’tri-element of property-effect-material’and esuggested that material composition within a CHM is the fundamentality to present its property. However, any single composition cannot express the CHMP integrally. Therefore, A variety of fingerprint technology have to be used to separate TCM material compositions. Identification of CHMP-markers through combination of these data can illustrate the overall CHMP. Comprehensive methods are need to analysis CHMP which is characterized through complicated combination of various compositions. As a modern analytical technology, TCM fingerprint can use to identify the complicated material compositions of herbs. We built CHMP recognized model based on TCM fingerprint to analysis the correlation between CHMP and material composition within a herb in quantity. There are20different kinds of TCM fingerprint of this’973program’. Based on understanding of TCM fingerprint and modern statistical pattern recognition we identify CHMP-markers. We built partial least squre discriminant analysis (PLS-DA) model to assess the performance of different levels of data. Following the above research meaning, we selected60classical CHMs in common use (their properties were determined in the light of ’Shen Nong’s Herbal Classic’,’Chinese materia medica’ and ’PRC codex’,30’cold’ and30’hot’ respectively). Here we utilized20kinds of fingerprint analysis technologies to obtain the data set. PLS-DA model were built to carry out the classification and prediction of CHMP based on different kinds of fingerprint data after preprocessing. We study the interrelationship between cold/hot property and material composition respectively based on inorganic elements, primary substances and secondary substances. First we built PLS-DA model based on single TCM fingerprint data to identify CHMP-markers and then select the features of this data. Then assemble these features at three levels (inorganic elements, primary substances and secondary substances) to build PLS-DA model. Last we combinate all these features to built PLS-DA model to show the interrelationship between the overall cold/hot property and material compositions. Analysis results:(1) The consistent homology rate of the complete dataset of PLS-DA model based on inorganic elements was75%(45/60). The cross-validation accuracy rate was54.33%(326/600). The consistent homology rate of the training dataset was70.83%(34/48). The forecast accuracy rate of the training dataset was66.67%(8/12). This model had poor performance in discrimination and prediction. The elements with VIP>1include:Al, Fe, Mg, V, Co, Zn, Sr and Cd. The consistent homology rate of the whole dataset of PLS-DA model based on selected elements was63.33%(38/60). The cross-validation accuracy rate was54.33%(326/600). The consistent homology rate of the training dataset was62.5%(30/48). The forecast accuracy rate of the training dataset was66.67%(8/12).(2) PLS-DA model based on single primary substance performed badly in discrimination and prediction. We selected variables with VIP>1to get a new data set. Then we combined these data sets to build a new PLS-DA model. The consistent homology rate of the whole dataset of this model was83.33%(50/60). The cross-validation accuracy rate was65.83%(359/600). The consistent homology rate of the training dataset was91.67%(44/48). The forecast accuracy rate of the training dataset was83.33%(10/12). This model performed better in prediction and discrimination.(3) PLS-DA model based on single fingerprint of secondary substance had poor performance in discrimination and prediction. We selected variables with VIP>1to combine a new data set to build a PLS-DA model. The consistent homology rate of the whole dataset of this model was83.33%(50/60). The cross-validation accuracy rate was65.83%(359/600). The consistent homology rate of the training dataset was91.67%(44/48). The forecast accuracy rate of the training dataset was83.33%(10/12). This model performed better in prediction and discrimination.(4) In PLS-DA model based on the assembled data of all kinds of data, the consistent homology rate of the whole dataset was100%, the consistent homology rate of the training dataset was100%, the forecast accuracy rate of the training dataset was75%and the cross-validation accuracy rate was78.17%. The model was stable and had better performance in discrimination.(5) The coefficients of PLS-DA model were used to identify CHMP-marker at different levels. The theory fingerprints had their own characteristics. The forcast fingerprint of a herb showed that cold or hot medicine contains’cold’material and ’hot’material composition at the same time. The CHMP is characterized through appropriate combination of various compositions in quality and quantity.Major conclusion:(1) Any single fingerprint can express only a small part of CHMP, so PLS-DA model based on single fingerprint data performanced poorly in discrimination and prediction.(2) PLS-DA models based on inorganic elements fingerprint data was not feasible to identify CHMP and to clarify the correlation between CHMP and different levels of materials. But PLS-DA model based on primary substances or secondary substances performed better in identifing CHMP-marker and forcasting the property of herbs.(3) PLS-DA model based on combination data set of inorganic elements, primary substances and secondary substances fingerprint had higher discrimiant and forcast accuracy rate.Major innovations:(1) Model based on combination data set of multi-dimensional and multi-data characteristic fingerprint was established and CHMP-markers were indentified according to the PLS discriminant coefficients.(2) We proved that any single fingerprint cannot express CHMP integrally. We combined the data of inorganic elements, primary substances and secondary substances as the whole data set of CHM, then PLS-DA model was established basd on the whole data set. This model can effectively identify and forecast CHMP.(3) Single TCM forcasting fingerprint can established based on PLS-DA model. The fingerprint can guide experimenters to separate substances of TCM.This research illustrated the relationship between CHMP and material composition in quantify, but we didn’t build the network between them. The next step of this research was mainly on the network. Meanwhile, the CHMP-markers we selected need to verified through animal experiments and clinical trials. |
PDF Full Text Request