基于多数据源协同学习方法的方剂组配规律研究

It is of great significance for the study of prescription matching rule to reduce the research and development cycle of prescription discovery, as well as shorten the related cost, and complete the prescription matching rules. Based on a comprehensive survey of current methods, technology and drawbacks of prescription matching rule, a drug matching rule schema is proposed based on multiple data sources collaborative learning. The study is focused on: 1) We research the quality assessment and selection of data source based on considering the complementarity and redundancy of data source, and evaluate the large-scale biological and pharmological data available for drug and disease, a complete and elaborated representations for these two entites are obtained. 2) From information and knowledge, with strategies such as large magin and matrix decomposition under entropy constraint, we construct a model to find out the drug matching rule under multiple sources cooperation by bringing the idea of collaborative recommend. 3) Based on this, function of drug and prescription, as well as description of diseases, can be applied to research the change of medical effect after distribution, so as to find out a new drug matching rule. The ideology and methodology studied in our project will definitely provide not only a theoretical basis for pharmacology, but also an important reference for mining new drug matching rule. It can also help the sorting and transmission of experience of famous specialists of Chinese medicine.

深入研究临床有效的方剂，充分挖掘新的药物组配规律，对于降低研发费用、缩短方剂组配研发周期、挖掘新的方剂组配规律具有重要的意义。在对现有药物组配规律研究技术的优缺点进行系统的归纳总结基础上，本项目提出一种基于多数据源协同学习方法发现药物组配规则的计算模型。主要内容：1）从数据源的互补性及冗余性出发，评估关于药物及疾病的海量多源异构描述信息，研究数据源的质量评估与选择方法，以充分有效地获取药物和疾病描述信息。2）分别从信息层面与知识层面出发，利用大间隔及熵约束矩阵分解等策略，引入协同推荐思想，构建多源协作下的药物组配规则发现模型。3）在此基础上，利用药物、方剂的功效信息及疾病描述信息，探讨药物组配后功效值的变化情况，从而挖掘新的药物组配规律。项目的研究思路和方案，将为方剂学提供理论依据，为新药品种研发提供重要的参考，辅助名老中医经验的整理和传承。

充分利用药物及疾病的多源描述信息，对发掘新的方剂药物组配规律、缩短方剂研发周期、缩减研发费用、完善方剂组配规律具有重要的意义。本项目提出了基于多数据源协同学习方法的方剂组配规律研究，主要内容为：（1）面向多源的药物及疾病描述信息的收集与整理，对单标记或多标记药物数据进行了特征选择等预处理。（2）针对多源数据，从数据源的互补性及冗余性出发，研究了数据源质量评估与选择方法。同时，分别从信息层面与知识层面出发，研究了信息层面下的多源数据融合模型及知识层面下的多源迁移学习。（3）引入协同过滤推荐模型，提出了多种协同过滤推荐模型，并根据药物对疾病有适应症和副作用等相关信息，设计了新的协同过滤模型以预测药物的新功效。本项目的研究为方剂药物的数据分析提供重要的技术支持，挖掘出的方剂药物新功效可以为方剂组配规律研究提供重要的数据参考。.针对上述研究内容，项目组基于预定研究方案开展工作，完成了原定研究计划和目标。本项目研究成果发表论文31篇，其中SCI收录期刊论文24篇，国内一级和核心期刊论文7篇，包括《IEEE TFS》、《Information Sciences》、《Applied Soft Computing》、《Knowledge Based Systems》、及《Neurocomputing》等重要学术期刊。培养硕士研究生数名。