Study On Identification Of Cold-Hot Nature And Visualization Of Characteristic Markers Based On Chinese Medicine Fingerprint

Objective: To correctly identify cold-hot nature of Chinese herbal medicines,and to explore the intrinsic law of cold-hot nature of Chinese herbal medicines plays an important role for Traditional Chinese Medicine treatment and medication.Relevant studies have pointed out that material components of Chinese herbal medicines are the basis for cold-hot nature of Chinese herbal medicines.It has been shown that material components of Chinese herbal medicines can be effectively characterized by Chinese medicine fingerprint.In this thesis,the ultraviolet spectrum in Chinese medicine fingerprint was used as an example to explore the establishment of identification model of cold-hot nature of Chinese herbal medicines.Meanwhile,the relationship between cold-hot characteristic markers and the identification results of cold-hot nature of Chinese herbal medicines was explored by the way of visualization of cold-hot characteristic markers of Chinese herbal medicines.We explored the relationship between material components and cold-hot nature of Chinese herbal medicines by the establishment of identification model and visualization of characteristic markers.Thus,it was further proved that Chinese herbal medicines with similar cold-hot nature have the same or similar material basis.Methods: In this thesis,the ultraviolet spectrum data of 61 Chinese herbal medicines in four different solvents(absolute ethanol,distilled water,chloroform and petroleum ether)were selected to establish a data set,and a cold-hot nature identification model based on e Xtreme Gradient Boosting(XGBoost)was created.The ultraviolet spectrum data of Chinese herbal medicines were converted into ultraviolet absorption curves,and two coldhot nature and deep learning identification models based on Res Net18 and Vision Transformer(Vi T)was proposed.For the problem of poor interpretability of complex machine learning models,this thesis used two interpretable machine learning algorithms,SHapley Additive ex Planations(SHAP)and Local Interpretable Model-Agnostic Explanations(LIME)to interpret the identification results of XGBoost model through visualization of cold-hot characteristic markers of Chinese herbal medicines.Results: In the aspect of cold-hot nature identification,among traditional machine learning methods,the identification model of cold-hot nature of Chinese herbal medicines based on XGBoost algorithm had the best performance.In the stability evaluation,the accuracy(ACC)and the area under ROC curve(AUC)of XGBoost model in petroleum ether solvent were 0.869 and 0.876 respectively,and the ACC in multi-solvent was 0.787.In the extrapolation evaluation,the ACC and AUC of XGBoost model in petroleum ether solvent were 0.838 and 0.841 respectively,and the ACC in multi-solvent was 0.790.Among deep learning methods,the ACC and AUC of the model built based on Res Net18 for the overall identification of Chinese herbal medicines were 0.738 and 0.745 respectively,and the ACC and AUC of the model built based on Vi T for the overall identification of Chinese herbal medicines were 0.918 and 0.958 respectively.In the aspect of visualization of cold-hot characteristic markers,cold Chinese herbal medicines had similarities in absorbance combination at ultraviolet wavelength 301,400 and 267 nm,while hot Chinese herbal medicines had similarities in absorbance combination at ultraviolet wavelength 301,400 and299 nm.The combination of the characteristic marker and value range of cold Chinese herbal medicines showed that the absorbance at 301 nm is greater than 0.00136 and less than or equal to 0.0076,the absorbance at 400 nm is greater than-0.00023,and the absorbance at267 nm is greater than 0.00542 and less than or equal to 0.037.The combination of the characteristic marker and value range of hot Chinese herbal medicines showed that the absorbance at 301 nm is greater than 0.0076,the absorbance at 400 nm is less than or equal to-0.0096,and the absorbance at 299 nm is greater than 0.0092.Conclusion: The results of cold-hot nature identification showed that the XGBoost model based on the ultraviolet spectrum table data of Chinese herbal medicines has the best identification performance among traditional machine learning methods.The Vi T model based on the ultraviolet absorption curve image data of Chinese herbal medicines can better identify cold-hot nature of Chinese herbal medicines among deep learning methods.The visualization results of cold-hot characteristic markers showed that cold and hot Chinese herbal medicines have the same or similar combination of the characteristic marker and value range respectively.This thesis further proved that Chinese herbal medicines with similar cold-hot nature have the same or similar material basis.