【作者】 汪涛；

【导师】 王亚东；

【作者基本信息】 哈尔滨工业大学 ， 计算机应用技术， 2020， 博士

【摘要】 随着基因组测序技术的快速发展以及测序成本的下降,大型人群基因组计划不断开展,基因组变异的发现数量得以空前增长。如何有效挖掘和认识基因组变异的分子功能,是功能基因组时代迫切需要解决的重要问题,对理解和研究疾病等重要性状的分子机理、发现药物靶点等具有重要意义。表达数量性状位点(expression Quantitative Trait Loci,eQTL)分析是解决该问题的重要途径,其通过在大量样本中分析基因组变异类型与基因表达强度之间的关系,可以从全基因组尺度挖掘基因组变异对基因的功能影响。然而,eQTL数据具有高噪声、数据不完整性、蕴涵复杂生物功能关系等特点,现有eQTL数据分析方法难以有效挖掘和揭示其中蕴涵的重要生物功能,迫切需要建立更加有效的数据处理与功能分析方法。本文围绕eQTL数据处理与功能分析方法开展研究,主要内容包括以下几个方面:第一,针对eQTL原始数据具有的多源异构性、普遍存在异常样本、普遍存在混淆因素带来的高噪声等问题,本文整合多种统计特征与模型,构建了一套涵盖数据质量分析、异常数据检测、混淆因素校正和数据归一化的eQTL原始数据清洗方法,解决了多源异构eQTL原始数据质量控制与归一化问题,有效降低人工干预、降低数据噪声、提高数据质量。第二,针对eQTL总结数据普遍存在缺失统计量的问题,本文提出基于多变量高斯分布的缺失eQTL数据推断方法。该方法基于变异之间的连锁不平衡关系,通过使用多变量高斯分布对eQTL统计量进行建模,用已知eQTL统计量对缺失eQTL统计量进行推断。通过全基因组片段化处理以及构建动态连锁不平衡关系矩阵,使该方法适用于并行分析,并有效提高缺失数据推断效率。该方法不需要原始基因型数据和原始基因表达数据,并适应于多种类型的QTL缺失数据推断。同时,该方法可以突破原始eQTL分析次要等位基因频率底线阈值,能够有效提高eQTL研究信号发现量。该方法有利于提高eQTL总结数据完整性、进一步增强eQTL研究的生物学发现。第三,针对eQTL与多基因间调控模式挖掘问题,本文提出eQTL调控网络构建与模体挖掘方法。该方法通过自适应置换检验策略以及基于广义帕累托分布拟合eQTL中介效应零分布,快速、高精度地计算eQTL中介效应,并整合eQTL二部网络与基于互信息的基因调控网络,构建eQTL调控网络。在此基础上,本文通过基于启发式子图同构判别的eQTL调控网络模体挖掘方法,降低子图同构判别计算空间,提高网络模体挖掘计算效率,挖掘eQTL与多基因间调控模式。该方法拓宽了传统基因调控模式研究思路,为eQTL与基因间的调控功能研究提供了新的方法支撑。第四,针对整合多组学数据进行生物分子网络中疾病模块挖掘问题,本文提出整合eQTL网络和多组学数据的疾病功能模块挖掘方法。该方法首先以eQTL网络作为桥梁,融合疾病表型与基因组变异关联数据(GWAS)、转录组基因表达与基因组变异关联数据(eQTL)、以及蛋白质相互作用数据(PPI),构建融合多组学数据信息的复杂生物分子网络。在此基础上,利用图表征学习模型对网络进行特征提取,利用层次聚类方法进行无偏向性的功能模块挖掘,利用疾病相关基因在模块中的富集水平对功能模块进行优先排序等,实现疾病功能模块挖掘。该方法有助于从系统生物学角度研究疾病分子通路,以及eQTL在疾病分子通路中的功能影响。更多还原

【Abstract】 With the rapid development of genome sequencing technology and the decline in sequencing costs,genome sequencing projects for large populations have been continuously launched,and the number of genome variants discovered has been unprecedentedly increasing.How to effectively decipher and understand the molecular function of genomic variation is an important problem that needs to be solved urgently in the functional genomic era.And it is of great significance for understanding and studying the molecular mechanism of diseases and other important traits,and discovering drug targets.Expression Quantitative Trait Loci(eQTL)analysis has become an important way to solve this problem.By analyzing the relationship between the genomic variation and the intensity of gene expression in a large number of samples,it is possible to investigate the functional impact of genomic variations on gene expression levels in genome-wide scale.However,eQTL data is characterized by high noise,incomplete data,and complex biological function relationships.The existing eQTL data analysis methods are not effective enough to mine and reveal the important biological functions contained in it.There is an urgent need to establish more effective data processing and functional analysis methods.This thesis focuses on eQTL data processing and functional analysis methods.The main contents include the following aspects:First,considering the characteristics of eQTL raw data such as multi-source heterogeneity,prevalence of abnormal samples,and high noise caused by confounding factors,we systematically built a set of eQTL raw data quality analysis and cleaning methods through the integration of multiple statistical models and machine learning methods,including data quality analysis,abnormal data detection,confounding factor adjustment,and data normalization.The method solves the problem of quality control and standardization of multi-source heterogeneous eQTL raw data,effectively reduces manual intervention,reduces data noise,and improves data quality.Second,considering the prevalence of missing statistics in eQTL summary datasets,this thesis proposes a method for inferring missing eQTL data based on multivariate Gaussian distribution.This method models eQTL statistics using multivariate Gaussian distribution based on the linkage disequilibrium relationships among DNA variations and infers missing eQTL statistics using known eQTL statistics.Through genome-wide fragmentation processing and construction of dynamic linkage disequilibrium relationship matrix,this method is suitable for parallel analysis,and can effectively improve efficiency of inferring missing data.This method does not require raw genotype data and raw gene expression data,and is suitable for inferring missing data of multipleQTL types.At the same time,this method can break through the lower bound threshold of minor allele frequency in original eQTL analysis,which can effectively increase the amount of eQTL signal discovery.This method is conducive to improving eQTL summary data integrity and further enhancing biological discoveries in eQTL studies.Third,aiming at solving the problem of mining regulatory patterns between eQTL and multiple genes,this thesis proposes a method for eQTL regulatory network construction and motif mining.This method uses the adaptive permutation testing strategy and fitting the null distribution of the eQTL mediation effects by the generalized Pareto distribution,to quickly and accurately calculate the eQTL mediation effects,and integrates eQTL mediation effects with the eQTL bipartite network and the gene regulation network based on mutual information to construct the eQTL regulatory network.On this basis,the thesis uses eQTL regulation network motif mining method based on heuristic subgraph isomorphism discrimination,to reduce the computational space of subgraph isomorphism discrimination,improve the computing efficiency of network motif mining,and discover the regulatory patterns between eQTL and multiple genes.This method broadens the research thinking of traditional gene regulatory patterns,and provides novel method support for the research of regulational functions between eQTL and genes.Fourth,aiming at mining disease modules from biomolecular networks integrated from multi-omics data,this thesis proposes a method of mining disease functional modules by integrating eQTL networks and multi-omics data.The method first uses the eQTL network as bridge to integrate disease phenotype and genomic variation association data(GWAS),transcriptome gene expression and genomic variation association data(eQTL),and protein-protein interaction data(PPI),and construct complex molecular network integrated from multi-omics data information.On this basis,graph representation learning model is used to extract the features of the network;hierarchical clustering method is used to unbiasedly detect the functional modules;and the enrichment level of the disease genes in the modules is used to prioritize the functional modules.This method is helpful to study disease molecular pathways,and the functional impact of eQTL in disease molecular pathways from perspective of system biology.更多还原