目的 以白及Bletillae Rhizoma为例，通过关联、决策和聚类等机器学习算法的联合使用进行数据挖掘研究，设计具有保胃、护肝双功能的保健食品配方并对其进行评价。方法 对现有具有保护胃黏膜或保护肝功能的保健食品批文，以及可用于治疗这2种疾病的中成药和方剂数据库中信息进行整理，先后使用关联规则算法（Apriori）、层次分析法（analytic hierarchy process，AHP）、SOM（self-organizing map）聚类、理想解法（technique for order preference by similarity to solution，TOPSIS）等关联、决策和聚类机器学习算法，挖掘其中高频原料药味的配方规律，结合现代功能活性及营养学研究成果，组成以白及为核心原料的具有配伍思想的保胃、护肝双功能保健食品配方并进行评价。结果 对各数据库进行原料药味频次统计和相应高频原料药味的关联性分析，发现高频原料药味之间极易产生较强的关联性。继而对数据库中出现的64个可用于保健食品的高频原料药进行2个层次共17个指标的AHP分析和加权后，得到除白及外加权值排在前5位的原料药味为甘草、陈皮、黄芪、茯苓和五味子，这些原料药可以考虑在组方时进行优先选择。SOM聚类显示所有高频原料可分为7类，其中在最优选配方原料药味与AHP分析结果权重排名前列的原料药味有极高的重叠。对原料进行保胃、护肝功能方面的文献检索，同时结合之前的数据分析结果，确定了15个原料进行君、臣、佐、使的配伍组合，最终设计了10个可能配方；并对其进行TOPSIS分析评价，其中排名前2位的配方黄芪、白及、五味子和甘草及黄芪、白及、陈皮和五味子的分值相近，均大于0.14，与其他配方评分相差较大。结论 在中医药基本理论的指导下结合各类机器学习算法，以核心药味白及为例建立了保胃、护肝双功能保健食品的配方设计和评价方法，为以后保健食品配方研发提供新的思路和方向。

Objective This study takes Baiji (Bletillae Rhizoma) as an example to design a health food formula with the protection of gastric mucosa and liver, furthermore develop an evaluation method. Some machine learning algorithms are taken into use in this study either, like association, decision, and clustering analysis. Methods Four databases from health food approval, Chinese patent medicines, and traditional Chinese medicine prescriptions with functions mentioned above were organized in order. Some machine learning algorithm, such as Apriori association rules algorithm, analytic hierarchy process (AHP), self-organizing map (SOM), and technique for order preference by similarity to solution (TOPSIS) was used in turn to find the regular relation among high-frequency ingredients which rooting from the databases. Combined with the function and nutrition research of all high-frequency ingredients, health food formula with the protection of gastric mucosa and liver, which full of thoughts of compatibility were designed and graded. Results Through the statistics of the frequency of all functional ingredients showed up and the correlation analysis of the corresponding high-frequency function ingredients, it was found that the high-frequency function ingredients were easy to have a stronger correlation with each other. Then 64 high-frequency functional ingredients from four databases above were analyzed by the AHP algorithm, including 17 indicators of two levels. The weight value showed that Gancao (Glycyrrhizae Radix et Rhizoma), Chenpi (Citri Reticulatae Pericarpium), Huangqi (Astragali Radix), Fuling (Poria), and Wuweizi (Schisandrae Chinensis Fructus) were in the top rank, which could be favorable evidence for the final formula. All functional ingredients were divided into seven groups in SOM clustering. The results indicated that the best choices group had a great common with the AHP top rank. Through literature search on the functions of protecting stomach and liver of ingredients, 15 functional ingredients were determined to be compatible with king, minister, adjuvant and enthral, combined with the previous data analysis results. Ten possible formula was comprised by 15 functional ingredients, with using TOPSIS analysis to provide the basis of the final formula. The grades showed in tie between formula composed with Astragali Radix, Bletillae Rhizoma, Schisandrae Chinensis Fructus, Glycyrrhizae Radix et Rhizoma and formula composed with Astragali Radix, Bletillae Rhizoma, Schisandrae Chinensis Fructus, Citri Reticulatae Pericarpium, which grades were all above 0.14 and have obvious difference with other formula. Conclusion Under the guidance of the basic theories of traditional Chinese medicine and combined with various machine learning algorithms, this study took the core medicine Bletillae Rhizoma as an example to establish the formula design and evaluation method of dual-function health food protecting stomach and liver, also provides a new idea and direction for the research and development of Chinese medicine prescription health food in the future.

R285.64