| 题名 | 基于生物网络的关联发现关键技术研究 |
| 作者 | 刘西 |
| 学位类别 | 工学博士 |
| 答辩日期 | 2012-05-29 |
| 授予单位 | 中国科学院研究生院 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 杨一平 |
| 关键词 | 本体 生物网络 贝叶斯回归 网络拓扑 致病基因 药物靶标 网络药理学 关联发现 Ontology Bio-network Bayesian Regression Network Topology Disease Gene Drug Target Network Pharmacology Association Discovery |
| 其他题名 | Research on Key Technology of Association Discovery Based on Bio-network |
| 学位专业 | 计算机应用技术 |
| 中文摘要 | 本文主要研究基于生物网络的关联发现技术，可用于解决在以“老药新用”为指导思想的药物研发过程中“如何确定目标疾病的候选药物”这一关键问题。该问题即等同于对候选药物的前瞻性预测问题，可具体分解为三个子问题： ①预测目标疾病的致病基因；②关联新的药物-靶标交互作用；③推理预测疾病-药物关联关系。为了验证本文所提理论及方案的可行性，开发了一套原型系统，进行了工程实践验证。主要研究内容包括以下几个方面： （1）本体知识库的构建 针对“老药新用”为指导思想的药物研发过程中“如何确定目标疾病的候选药物”这一关键问题，结合本体的相关理论，构建了三类生物本体模型，分别是疾病本体、药物本体、蛋白质本体，并针对目前权威的6 个开放生物医学数据库，给出了一个基于本体的数据集成框架，构建了对应的三类生物本体知识库，为后续的关联发现研究提供了统一且通用的数据基础和知识服务。 （2）预测目标疾病的致病基因 针对目标疾病的致病基因预测问题，根据网络拓扑特征分类预测和贝叶斯回归分析预测的方法特点，提出了一种联合的致病基因预测算法。以生物本体知识库为数据基础，构建疾病-基因网络和蛋白质交互网络，分析候选基因在PPI （蛋白质交互）网络中的拓扑特性，根据已知致病基因的拓扑特征训练分类器，再对候选基因进行分类，得到潜在致病基因的初步预测结果，再将此预测结果作为先验知识，带入后续的贝叶斯回归预测模型中，并引入贝叶斯因子的概念用于度量候选基因与目标疾病的关联性大小，依此关联指标最终确定目标疾病的潜在致病基因。 （3）关联新的药物-靶标交互作用 为解决新的药物-靶标交互作用预测问题，针对前人基于二分图建模的预测方法中存在的不足，提出了一种新的基于二分图评价模型的改进预测算法。该算法根据已知的药物-靶标交互作用构建二分图网络，分别从药物化学结构信息和药物疗效信息两个角度来综合刻画药物-靶标作用对的关联性评价模型，再依此模型预测新的药物-靶标蛋白质交互作用，即预测新靶标。 |
| 英文摘要 | This paper is focused on the research of association discovery technology based on biological networks. The research is used to resolve the key problem of "how to determine the candidate drugs of target disease" in procedure of drug research and development, that takes "finding new utilities of old drugs" as its basic principle. This issue equals predicting candidate drugs, and includes three sub-issues: 1. Predict pathogenic gene of target disease; 2. Find new interactive connections among drugs and targets; 3. Derive associations between drug and disease. To test and verify the theory in this paper, a prototype system has been developed. The main content of the research is as followed: (1) Construction of ontological knowledge base To solve the key issue we mention above, three types of biological ontology model have been constructed: disease ontology, drug ontology and protein ontology, based on related theory of the ontology. And we analyze six authority opening biological databases and develop a basic comprehensive data framework about ontology. It includes three types of corresponding ontological knowledge base and can provide a unified and common data base for the subsequent research. (2) Predicting pathogenic gene of target disease To predict disease-causing genes of target disease, we propose a fusion prediction method based on the network topology properties and Bayesian regression analysis. We take biology ontology knowledge base as foundation to construct gene-disease network and protein-protein interaction network. Then, we analyze the topological features of candidate genes in PPI network. Based on these topological features of known disease genes, we train a classifier to classify unknown genes and predict the initial possible disease genes. This result can be used as priori knowledge in the following Bayesian regression model we construct. Finally, we introduce the concept of Bayes factor to measure the degree of relevance and confirm pathogenic gene of target disease by checking this relevance. (3) Discovery of new association between drug and target To resolve this problem, a predicting algorithm based on the improved bipartite graph evaluation model is proposed, according to the deficiencies of previous research. This algorithm constructs a bipartite graph network based on known interactions between drugs and target proteins. The evaluation model of association between drugs and target proteins is established based on two aspects of ... |
| 语种 | 中文 |
| 其他标识符 | 200918014629091 |
| 内容类型 | 学位论文 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/6445]  |
| 专题 | 毕业生_博士学位论文 |
| 推荐引用方式 GB/T 7714 | 刘西. 基于生物网络的关联发现关键技术研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2012. |
| 个性服务 |
| 查看访问统计 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。
修改评论