基于复杂网络的药物副作用分子机理分析研究

发布时间：2018-04-05 01:25

本文选题：药物副作用　切入点：BGLL　出处：《北京交通大学》2017年硕士论文

【摘要】：药物副作用是病人在治疗的过程中按照正常的药物剂量进行诊断、预防、治疗某种疾病所出现的与治疗目的无关的反应,一般会引起患者的不适和痛苦。美国的一项关于药物副作用的研究表明严重的药物副作用是引起人类死亡的第四大原因,每年会引起1,000,000人死亡。因此,药物副作用逐渐成为公共健康的主要问题,其不仅是药物研发失败的主要原因,同时也是新药研究与投产的主要阻力。目前,对于药物副作用的研究主要集中在如下三个方面:一、利用药物所作用的靶点来预测药物副作用;二、基于药物的化学结构预测药物副作用;三、利用数据挖掘技术从文献库中挖掘药物和副作用的关系来丰富和完善现有的数据库。然而,已有药物副作用的研究仅仅考虑药物自身特性,并未探究疾病、基因、症状与药物副作用之间的关系。为此,本文主要研究内容有以下三点。首先,整合来自不同数据库中的数据,形成副作用与疾病、基因、症状等关系数据,建立基于基因的疾病网络,基于副作用的疾病网络,基于症状的疾病网络等。为了探索副作用和疾病之间的关系,采用BGLL和BigCLAM(Cluster Affiliation Model for Big Networks,简称BigCLAM)两种复杂网络社团划分算法对基于副作用的疾病网络进行社团划分。两种社团划分算法所得结果均具有较高的模块度,且所得模块具有高度的一致性,由此表明疾病和副作用之间存在相关关系。其次,从富集分析和基因一致性分析两方面验证疾病和副作用之间的关系。通过对高相似度的疾病网络进行网络社团分析,得到441个模块,采用富集分析方法对疾病模块和疾病分类进行分析,实验表明同一模块中的疾病富集到同一疾病类别。基因一致性分析是利用基因疾病网络数据和副作用疾病网络数据,计算两个网络在实验条件和随机条件下的重叠特性。结果表明实验条件下网络的重叠特性远远高于随机条件,由此验证了疾病与副作用之间的关系。之后采用支持向量机和简单逻辑回归分类方法,以副作用为特征对疾病分类(文中以C01类疾病为例)。不同分类方法均可得到较高的精准率、召回率和AUC(AreaUnderroc Curve,简称AUC),结果表明部分副作用可能是疾病本身的属性。最后,从副作用的类别出发得到副作用主要属于T047疾病类别和T184症状类别。对于T047副作用的网络拓扑特性,如度分布进行相关研究发现同一节点所关联的疾病多数属于同一类别,结果表明药物治疗一类疾病中的一种疾病同时会引起该类疾病中的其它疾病。通过对T184副作用基因网络的构建和基因功能的分析,结果表明同一模块中的基因具有相同的功能特性,同时与某一类别的症状相关联,得到决定症状类副作用的相关基因。通过对以副作用为中心的不同分子网络的研究,对副作用与疾病、基因、症状之间的关系有了更深一步的认识,为研究副作用的微观机理提供了帮助。
[Abstract]:Drug side effect is the patient in the course of treatment according to the normal dosage of drugs for diagnosis, prevention, treatment of a disease and the treatment of non-response to the purpose of treatment, generally will cause discomfort and pain of the patient.A U.S. study of drug side effects shows that severe side effects are the fourth leading cause of death in humans, causing 1000000 deaths a year.Therefore, side effects of drugs have gradually become the main problem of public health, which is not only the main reason of drug development failure, but also the main resistance of new drug research and production.At present, the research on drug side effects is mainly focused on the following three aspects: first, using the targets of drugs to predict drug side effects; second, predicting drug side effects based on the chemical structure of drugs; third,The data mining technology is used to mine the relationship between drugs and side effects from the document library to enrich and perfect the existing database.However, studies of drug side effects have only considered the characteristics of the drug itself and have not explored the relationship between disease, genes, symptoms and drug side effects.Therefore, the main content of this paper has the following three points.Firstly, we integrate the data from different databases to form the relationship data between side effects and diseases, genes, symptoms and so on, and establish the network of diseases based on genes, disease networks based on side effects, disease networks based on symptoms and so on.In order to explore the relationship between side effects and diseases, BGLL and BigCLAM(Cluster Affiliation Model for Big Networks (BigCLAM) are used to divide disease networks based on side effects.The results of the two algorithms have a high degree of modularity and a high degree of consistency, which indicates that there is a correlation between disease and side effects.Secondly, the relationship between disease and side effects was verified by enrichment analysis and gene consistency analysis.Through the network community analysis of the disease network with high similarity, 441 modules were obtained, and the disease module and disease classification were analyzed by the enrichment analysis method. The experiment shows that the disease in the same module is enriched to the same disease category.Gene consistency analysis is based on genetic disease network data and side-effect disease network data to calculate the overlapping characteristics of the two networks under both experimental and random conditions.The results show that the overlap property of the network under experimental conditions is much higher than that of random conditions, which verifies the relationship between disease and side effects.Then support vector machine (SVM) and simple logic regression were used to classify diseases with side effects (C01) as an example.Different classification methods can obtain higher precision, recall rate and AUC(AreaUnderroc current, the results show that some side effects may be the attribute of the disease itself.Finally, from the category of side effects, we found that side effects mainly belong to the category of T 047 disease and the category of symptoms of T 184.For the network topology characteristics of T047 side effects, such as degree distribution, it is found that most diseases associated with the same node belong to the same category.The results show that one disease in one disease can cause other diseases in the same time.Through the construction of T184 side effect gene network and the analysis of gene function, the results show that the genes in the same module have the same functional characteristics, and at the same time associated with a certain type of symptoms, the related genes that determine the symptom side effects are obtained.Through the study of different molecular networks centered on side effects, a deeper understanding of the relationship between side effects and diseases, genes, and symptoms has been obtained, which will help to study the microcosmic mechanism of side effects.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R96;O157.5

【参考文献】