方剂间相似性的量化表征方法及模型应用

Quantitative characterization method and applicated model of similarity between formulas

方剂数据以每年10亿条的规模增加,针对庞大方剂数据如何提高研究结果的准确度、效度、降低资源消耗,是当前中医药数据挖掘领域关键问题之一。前期研究表明,对方剂数据进行合理的分类、聚类及实体消歧具有重要的应用价值。然而,如何对方剂间的相似性进行量化表征是解决该问题的关键技术。本研究通过对现有方剂相似度的相关研究进行系统复习,梳理出基于字符串、集合、向量距离、概率分布的量化表征方法及应用模型,提出了基于复杂网络和深度学习技术对方剂相似度量化的研究思路,为基于方剂相似度挖掘核心方药、药物配伍、药症关联规律研究提供了方法学参考。

The prescription data is increasing by 1 billion pieces every year.How to improve the accuracy,validity,and reduce resource consumption of the research results based on the huge prescription data is one of the key issues in the field of TCM data mining.Preliminary studies have shown that reasonable classification,clustering,and entity disambiguation of formula data have important application value.However,how to quantify the similarity between the counterparts is the key technology to solve this problem.By systematically reviewing the related research on the formula similarity,we extract the quantitative methods and applicated models based on string,set,vector distance,and probability distribution.We propose the idea of quantifying the formula similarity based on complex networks and deep learning technologies.The quantified research provide a methodological reference for mining core prescriptions,drug compatibility,and drug-symptom association rules based on formula similarity.