基于XGBoost和SHAP的中药寒热药性识别及寒热特征标记可视化研究  

Study on Identification of Cold-Hot Nature and Visualization of Cold-Hot Characteristic Markers of Chinese Herbal Medicines Based on XCBoost and SHAP

作　　者：张喜科 赵文华[1] 付先军 马志庆[1] 魏国辉 Zhang Xike;Zhao Wenhua;Fu Xianjun;Ma Zhiqing;Wei Guohui(College of Intelligence and Information Engineering,Shandong University of Traditional Chinese Medicine,Jinan 250355,China;Key Laboratory of Theory of Traditional Chinese Medicine,Ministry of Education of China,Shandong University of Traditional Chinese Medicine,Jinan 250355,China)

机构地区：[1]山东中医药大学智能与信息工程学院,济南250355 [2]山东中医药大学中医药经典理论教育部重点实验室,济南250355

出　　处：《世界科学技术-中医药现代化》2022年第12期4984-4993,共10页Modernization of Traditional Chinese Medicine and Materia Medica-World Science and Technology

基　　金：国家自然科学基金委员会面上项目(81473369):基于“性-构”关系的中药成分寒热药性评价,负责人:付先军。

摘　　要：目的运用集成学习中的XGBoost算法探索构建中药寒热药性识别模型,采用SHAP进一步探索中药寒热特征标记与寒热药性识别结果的关系。方法利用61味中药的紫外光谱数据集,运用集成学习中的XGBoost算法,构建中药寒热药性识别模型,通过稳定性评价和外推评价等方式评估模型的性能,采用SHAP对中药寒热特征标记进行可视化。结果模型评价方面,在稳定性比较中,本文模型在单溶剂下的ACC和AUC分别为0.852、0.865,多溶剂下的ACC为0.770;在外推比较中,本文模型在单溶剂下的ACC和AUC分别为0.810、0.828,多溶剂下的ACC为0.786;可视化方面,寒性中药在紫外波长400和267 nm的吸收度具有相似性,而热性中药在紫外波长400、299和301 nm的吸收度具有相似性。结论与经典模型相比,本文模型对中药寒热药性具有更好的识别能力。具有相似的紫外光谱的中药,其药性同样具有相似性,初步证明了物质成分相似的中药具有相似的药性。Objective The XGBoost algorithm in ensemble learning was used to explore the establishment of the identification model of cold-hot nature of Chinese herbal medicines,and the relationship between cold-hot characteristic markers and the identification results of cold-hot nature of Chinese herbal medicines was further explored by SHAP.Methods Using the ultraviolet spectrum data set of 61 Chinese herbal medicines,the XGBoost algorithm in ensemble learning was used to construct the identification model of cold-hot nature of Chinese herbal medicines,the performance of the model was evaluated by stability evaluation and extrapolation evaluation,and the cold-hot characteristic markers of Chinese herbal medicines were visualized by SHAP.Results In terms of model evaluation,in the stability comparison,the ACC and AUC of this model in single solvent were 0.852 and 0.865 respectively;The ACC under multi-solvent was 0.770.In the extrapolation comparison,the ACC and AUC of this model in single solvent were 0.810 and 0.828 respectively;The ACC under multi-solvent was 0.786.In terms of visualization,the absorption of cold Chinese herbal medicines at UV wavelengths of 400 and 267 nm was similar,while that of hot Chinese herbal medicines at UV wavelengths of 400,299 and 301 nm was similar.Conclusion Compared with the classical model,the model in this paper has better identification ability for cold-hot nature of Chinese herbal medicines.Chinese herbal medicines with similar UV spectrum have similar nature,it is preliminarily proved that Chinese herbal medicines with similar material components have similar nature.

关 键 词：中药寒热药性 中药寒热特征标记 集成学习 XGBoost SHAP 可视化 

分 类 号：R285.1[医药卫生—中药学]