基于多特征融合的药物作用模式统计分析方法研究

发布时间：2019-04-26 08:07

【摘要】：药物作用模式(MoA)研究对于药物设计研发、预测药物未知副作用以及指导用药具有重要意义。伴随着新药产率逐年呈下降趋势,预测已有药物的潜在作用模式被认为是解决新药开发高投入低产出的有效方法之一。对已知药物的作用模式进行分析,可以帮助发现药物可能的潜在应用和未知的副作用。本文以美国食品与药品监督管理局(FDA)批准的药物为研究对象,通过统计分析与数据挖掘方法,分析各药物信息与作用模式的关系,挖掘不同作用模式之间的药物信息的差异性,结合多特征融合的概率集成方法预测潜在的药物作用模式,为临床指导用药提供一定的参考依据。主要内容如下:1.提出了一种利用药物的化学结构信息、生物特性、药理学特性以及药物副作用来系统分析评估药物作用模式的方法。该方法主要使用了药物信息为特征,通过统计分析方法,分析不同药物作用模式下各种药物特性信息的差异性及富集情况,以及验证这些特性特征是否可以较好的区分不同的药物作用模式。实验结果表明,不同药物作用模式类别下药物的各个信息均具有显著性差异,且这些药物特性信息可以较好的区分不同的药物作用模式。2.提出了一种基于多特征融合的概率集成方法构建药物作用模式网络挖掘潜在的药物作用模式。该方法基于贝叶斯网络模型理论对药物的四个特性的相似性特征进行融合,进而结合概率论知识构建网络挖掘潜在的药物作用模式。实验结果表明,多特征融合对模型预测性能确实有效,通过对比其他四种不同的机器学习模型对药物作用模式的预测性能发现,概率集成方法预测精度最高,鲁棒性较好,而且可以成功预测出一些潜在的药物作用模式。3.构建了一个在线药物作用模式类别的药理学数据库"MoABank"。该数据库可以提供较为全面的药物作用模式类别及药物的靶标、通路及副作用等信息,以及本文模型的分析与预测结果。
[Abstract]:The study of drug action model (MoA) is of great significance for drug design and development, prediction of unknown side effects of drugs and guidance of drug use. As the yield of new drugs decreases year by year, predicting the potential action models of existing drugs is considered to be one of the effective methods to solve the problem of high input and low output in the development of new drugs. An analysis of the patterns of action of known drugs can help to identify potential applications and unknown side effects of drugs. In this paper, the drug approved by the Food and Drug Administration of the United States of America (FDA) as the research object, through statistical analysis and data mining methods, to analyze the relationship between drug information and action patterns. Mining the difference of drug information among different action patterns, combining the probability integration method of multi-feature fusion to predict the potential drug action patterns, which provides a certain reference basis for clinical guidance of drug use. The main contents are as follows: 1. This paper presents a method to systematically analyze and evaluate drug action patterns by using chemical structure information, biological characteristics, pharmacological properties and side effects of drugs. This method mainly uses the drug information as the characteristic, through the statistical analysis method, analyzes the difference and the enrichment of the various drug characteristic information under the different drug action mode. And verify whether these characteristics can better distinguish the different drug action patterns. The experimental results show that there are significant differences in the information of different drug types under different drug action modes, and these information of drug characteristics can be used to distinguish different drug action patterns. 2. A probability ensemble method based on multi-feature fusion is proposed to construct drug action patterns network mining potential drug action patterns. Based on Bayesian network model theory, the similarity characteristics of the four characteristics of drugs are fused, and then the potential drug action patterns are constructed by combining the knowledge of probability theory. The experimental results show that the multi-feature fusion is effective for the prediction performance of the model. By comparing the prediction performance of the other four different machine learning models to the drug action model, the probability ensemble method has the highest prediction accuracy and better robustness. And some potential drug action patterns can be predicted successfully. 3. A pharmacology database "MoABank" for online drug action model categories was constructed. The database can provide more comprehensive information about the types of drug action patterns, targets, pathways and side effects of drugs, as well as the results of analysis and prediction of the model in this paper.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R91

【参考文献】