基于网络结构推理和拓扑相似性的致病基因预测算法

发布时间：2018-05-04 19:27

本文选题：网络结构推理 + 二部图　；参考：《安徽大学》2017年硕士论文

【摘要】：现阶段,研究者们发现几乎所有的疾病都与基因有一定的关联,如何迅速地找到疾病的致病基因对所有研究者们来说是一项很大的挑战。早期,研究者大多使用传统生物实验的方法来进行致病基因的预测,比如连锁分析和关联研究。近几年来,研究者们开始大量使用生物信息学中的计算方法,并通过基于网络结构方法来解决该难题。然而,大多数方法在推理过程中仅仅使用了网络结构中的局部网络信息,并且仅限于推理单一基因的关联。此外,这些方法很少或几乎没有考虑到疾病与基因关联网络中的网络拓扑相似性。本文结合生物信息学和数据挖掘领域中的相关知识,使用了 OMIM(Online MendelianInheritance inMan,OMIM)数据库中的203个疾病数据作为实验的数据集来进行致病基因的预测。本文主要的研究内容和创新工作概述如下:1.为了预测致病基因,本文首次引入二部图网络结构推理(Network-Based Inference,NBI)算法。具体的步骤如下:首先,构建疾病和基因的关联二部图网络;其次,利用物质热扩散原理将基因上的初始化资源扩散到疾病上;最后,再将疾病上的资源扩散到基因上。上述操作结束后将会获取到候选基因的资源向量,并对资源向量按值大小排序进而预测潜在致病基因。在使用OMIM数据集中的203个疾病数据作为最终的实验数据,通过留一交叉验证实验来证明该算法的有效性。2.针对二部图网络结构推理的算法,提出了一种基于网络拓扑相似性的改进算法,称为 INBI(Improved Network-Based Inference,INBI)算法。首先,分别计算网络拓扑相似性,本文通过使用高斯核函数来计算基因网络的拓扑性和疾病网络的拓扑相似性;其次,分别构建疾病邻接矩阵和基因邻接矩阵;然后,预测致病基因的相关性分数;最后,根据相关性分数进行排序从而预测潜在的致病基因。在进行对比实验结果的分析后,可以得出INBI算法具有优越的性能。
[Abstract]:At present, researchers have found that almost all diseases are related to genes, and how to quickly find the disease-causing genes is a great challenge to all researchers. In the early days, researchers mostly used traditional biological experiments to predict pathogenic genes, such as linkage analysis and association studies. In recent years, researchers have begun to use a large number of computational methods in bioinformatics and solve the problem based on network structure. However, most of the methods only use the local network information in the network structure in the reasoning process, and only infer the association of a single gene. In addition, these methods rarely or hardly take into account the network topology similarity between disease and gene associated networks. Combining the related knowledge in the field of bioinformatics and data mining, 203 disease data in the OMIM(Online MendelianInheritance in Manmim database are used as experimental data sets to predict the pathogenic genes. The main research contents and innovative work of this paper are summarized as follows: 1. In order to predict the pathogenic genes, the bipartite graph network structure inference (NBI) algorithm is introduced for the first time in this paper. The specific steps are as follows: first, construct the bipartite network of disease and gene association; secondly, use the principle of material thermal diffusion to spread the initial resources on the gene to disease; finally, spread the disease resources to the gene. The resource vector of candidate gene will be obtained after the operation and the resource vector will be sorted according to the value to predict the potential pathogenic gene. Using 203 disease data in the OMIM dataset as the final experimental data, the validity of the algorithm is proved by a cross-validation experiment. In this paper, an improved algorithm based on network topology similarity is proposed for bipartite graph network structure reasoning, which is called INBI(Improved Network-Based reference algorithm. Firstly, the topological similarity of the network is calculated, and the topological similarity of the gene network and the disease network is calculated by using Gao Si kernel function. Secondly, the disease adjacency matrix and the gene adjacency matrix are constructed respectively. The correlation score of pathogenic gene is predicted. Finally, the potential pathogenic gene is predicted by sequencing according to the correlation score. After comparing the experimental results, it can be concluded that the INBI algorithm has superior performance.
【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R3416;TP311.13

【相似文献】