目的 基于西洋参“保健食品原料”与“中药材”的双重身份，以功能为导向，融合算法分析，实现西洋参保健食品组方规律挖掘与组方设计研究。方法 依据功能导向-算法集成，收集西洋参保健食品数据与临床应用数据，借助BDP平台、自然语言处理与信息检索（natural language processing&information retrieval sharing platform，NLPIR）共享平台实现数据处理与信息挖掘，综合分析具有潜在开发价值的保健食品功能；以选定功能为导向，收集西洋参保健食品、中成药、方剂的组方信息并建立数据库，统计高频药味及类别，采用Apriori关联规则算法、层次分析法（analytic hierarchy process，AHP）-熵权-自组织特征映射（self-organizing maps，SOM）联用的赋权聚类算法（主观与客观算法联用）、逼近理想解排序（technique for order preference by similarity to an ideal solution，TOPSIS）综合评价法等机器学习算法提取样本数据特征并分析总结其内在规律，实现组方设计。结果 综合西洋参保健食品功能与临床疾病和证型双方信息的分析结果，以“有助于增强免疫力”“对化学性肝损伤有辅助保护功能”双功能为导向，收集组方信息并建库。数据库中高频保健食品原料药味以补虚药居多，经赋权聚类筛选得到黄芪、麦冬、茯苓、五味子、枸杞子、三七、葛根、甘草共8味候选保健食品原料药味。结合药味关联分析及药对组合情况，进一步配伍组成备选组方，经评分优选得到保健食品组方为三七、西洋参、葛根、五味子、枸杞子。结论 针对中药保健食品的中医药背景，依托大数据平台分析挖掘保健食品功能，借助机器学习算法抽提组方配伍规律，实现功能导向-算法集成西洋参双功能保健食品组方设计，或可为中药保健食品组方设计的功能导向化、组方计算化提供新思路。

Objective Based on the dual-identity of Xiyangshen (Panacis Quinquefolii Radix) as health food functional ingredients and traditional Chinese medicine, the functional orientation and algorithm analysis were combined to realize the mining of health food formula rule and formula design of Panacis Quinquefolii Radix. Methods According to the function orientation-algorithm integration, the data of health food and clinical application of Panacis Quinquefolii Radix were collected and processed by business data platform (BDP) and natural language processing and information retrieval (NLPIR) sharing platform to analyze the health food functions with potential development value comprehensively through the information mining. Guided by selected functions, the formula information of Panacis Quinquefolii Radix health food, Chinese patent medicine and prescription was collected to establish a database. The high-frequency health food functional ingredients and categories were counted. The characteristics of data were extracted and the formula internal rules were analyzed and summarized through machine learning algorithms as Apriori algorithm, analytic hierarchy process (AHP)-entropy-self-organizing maps (SOM) algorithm (subjective and objective algorithms combined) and technique for order preference by similarity to an ideal solution (TOPSIS) to achieve the formula design. Results Based on the analysis results of health food functions and clinical disease and syndrome types, the formula information was collected and the database was established with the orientation of "contributing to boost immunity" and "auxiliary protective function on chemical liver injury". In the database, most of the high-frequency health food ingredients were tonifing deficiency herbs. Eight candidate health food ingredients including Huangqi (Astragali Radix), Maidong (Ophiopogonis Radix), Fuling (Poria), Wuweizi (Schisandrae Chinensis Fructus), Gouqizi (Lycii Fructus), Sanqi (Notoginseng RadixetRhizoma), Gegen (Puerariae Lobatae Radix) and Gancao (Glycyrrhizae Radix et Rhizome) were obtained by weighted clustering. Combined with the herbs association and herb pair analysis, the alternative prescriptions were further formed. The health food prescription optimized by scoring was consisted of Notoginseng RadixetRhizoma, Panacis Quinquefolii Radix, Puerariae Lobatae Radix, Schisandrae Chinensis Fructus and Lycii Fructus. Conclusion According to the traditional Chinese medicine background of Chinese materia medica health food, relying on the big data analysis platform for the health food functions, the prescription rules were summarized and extracted by the machine learning algorithms. The realization of the function orientation-algorithm integration ofPanacis Quinquefolii Radixhealth food formula rule mining and dual-function health food formula design may provide a new development idea for the function-oriented and formula-calculated of Chinese materia medica health food formula design.

R283.21；TS218；TP312

桂林市技术创新引导计划(2020010902);河南省重大科技专项(211110310100)