【作者】 刘伟；

【导师】 郭安源；

【作者基本信息】 华中科技大学 ， 生物信息技术， 2017， 博士

【摘要】 生命体在不同发育时期和不同组织部位,所表达基因的种类和数量不同,从而导致细胞分化、个体发育及不同组织器官的不同结构和生理功能。转录因子(Transcription factor,TF)是这种在特定的时间和空间条件下,基因选择性表达的重要调控因子,它能结合到特定DNA序列上发挥基因转录调控作用。获得转录因子的靶基因是转录因子调控研究的基础,不同组织条件下的转录因子靶基因对该转录因子的调控研究也有重要参考意义。本项目中我们收集了人类转录因子的多种公共数据,采用改进的预测方法进行靶基因预测,综合考虑和分析了表观修饰状态对转录因子结合的影响,预测了相对可靠的人类转录因子靶基因,并构建了开源的人类转录因子靶基因数据库,同时分析了转录因子的细胞系特异性调控、转录因子间的协作调控和转录因子对IncRNA的调控等,取得了如下研究成果。第一、我们获得了较大规模的人类转录因子靶基因数据,并构建了相应数据库。我们使用了 ChIP-Seq分析和转录因子结合位点(Transcription Factor Binding Sites,TFBS)扫描两种方法预测转录因子的靶基因。第一种方法是基于ChIP-Seq高通量测序的实验数据。首先,我们从ENCODE和NCBI等数据库收集了 488个人类转录因子的3231组ChIP-Seq数据,预测该转录因子在基因组上的结合峰(Peak)。然后根据前1000个结合峰预测并过滤出至多5个可信的模体(Motif),再在所有结合峰中扫描可信模体获得转录因子的结合位点(TFBS),并根据TFBS与转录起始位点的距离,采用指数衰减的BETA模型量化转录因子对靶基因的调控能力,得到转录因子靶基因。最后,汇总每个转录因子在多个样本中的靶基因预测结果及考虑靶基因的表观修饰状态,进一步过滤得到转录因子整合的靶基因预测结果,转录因子靶基因的中位数是342。第二种方法使用转录因子已知的结合模体预测靶基因。我们从TRANSFAC、JASPAR、HOCOMOCO等转录因子数据库收集了 699个转录因子的2737个模体的位置权重矩阵(Position Weight Matrix,PWM),通过扫描人类、小鼠、大鼠的基因组保守区域预测转录因子的潜在靶基因。基于上述结果,我们构建了人类转录因子靶基因的数据库 hTFtarget(http://bioinfo.life.hust.edu.cn/hTFtarget),方便用户查询和使用。用户可以通过多种查询方式方便地获取预测结果,包括查询转录因子靶基因和查询靶基因的转录因子,也可以进行联合查询、批量查询或者对预测结果进行筛选。第二、我们根据预测结果对转录因子的调控机理进行了探索。首先,我们探索了转录因子结合位点与表观修饰状态的关联,发现在多数情况下转录因子排序靠前的结合峰倾向于结合到活化的转录起始位点及其扩展区域附近,并且同一转录因子的多个样本或者同一家族的转录因子,它们的结合峰倾向于富集到相似的表观修饰状态。其次,我们分析了 14个转录因子在10个不同细胞系中普遍性调控和特异性调控的靶基因,并根据这些靶基因功能富集的结果讨论了这些转录因子在一般状态下和特定组织中的功能。最后,我们使用机器学习的方法分析了各细胞系中转录因子间的协同作用,得到了伴侣因子对转录因子的相对重要系数(RelativeImportance,RI),定量地描述了转录因子间协同作用的强弱。第三,我们分析了转录因子对lncRNA的调控作用,及SNP对转录因子结合的影响。首先,我们分析了转录因子对不同类型基因的调控情况,发现转录因子更倾向于调控蛋白编码基因和lncRNA基因。其次,我们预测得到了可能造成调控关系失效的9,815,083个SNP位点,其中231,558个SNP位于lncRNA基因的启动子区域,这些SNP位点可能导致转录因子对lncRNA调控的丢失。最后,我们将这些导致调控关系失效的SNP位点,与我们另一个收集整理lncRNA与SNP关系的数据库lncRNASNP中的lncRNA-SNP数据进行了比较,其中68,597个SNP位于lncRNASNP中。本研究采用系统的方法预测了人类转录因子的靶基因,构建了人类转录因子靶基因数据库,并探索研究了转录因子的特异调控、协同调控和对lncRNA的调控等,这些数据和研究为转录因子的基因表达调控研究提供了可靠的数据资源,并加深了对其调控机制的理解。更多还原

【Abstract】 Gene expression patterns changed in different tissues and developmental stages,leading to cell differentiation,ontogenesis,and different structures and functions of organs.Transcription factors(TFs)are key regulators in gene expression,they play crucial roles in gene expression both spatially and temporally.TFs are able to recognize and bind to special DNA sequence,to promote or block the specific gene expression,therefore regulate the expression of specific target genes.Identifying the targets of a TF could lay foundation for the study of gene regulation.In this study,we collected human TF ChIP-Seq data from a variety of public resources,and customized an analysis workflow to detect reliable TF targets with taking epigenomic states into account.The predictions are available in an open access database.Meanwhile,we also analyzed the tissue-specific transcriptional regulation,TF co-association,and the TFs that regulate long non-coding RNAs(lncRNA)and got the following results.Firstly,we obtained a large-scale of human TF targets prediction results,and built an open-source database hTFtarget.Two different strategies were applied to predict TF targets,including ChIP-Seq analysis and transcription factor binding sites(TFBS)scanning.The first approach is based on ChIP-Seq high throughput sequencing data.Firs of all,we collected 3,231 datasets of 488 human TFs from public database such as ENCODE and NCBI,and detected their binding peaks on human genome.Next,we extracted less than 5 motifs from the top 1000 peak sequences,and scanning TFBS with these motifs among all the peaks.Then we quantized the regulatory ability of TF to target gene with the distance from transcription start site(TSS)using BETA,a scoring model with exponentially decaying.Finally,to provide an integrated targets prediction,targets prediction result of individual dataset is collected and filtered according to the epigenomic states of histone modification.As a result,the median targets of each TF is 342.The second approach is predicting TF targets with known motifs.We collected 2,737 position weight matrix(PWM)of 699 TFs from TRANSFAC,JASPAR and HOCOMOCO,and scanning these motifs in conserved regions of human,rat and mouse genome to identify potential targets of these TFs.Based on these results we achieved,we built a comprehensive human transcription factor target database named hTFtarget(http://bioinfo.life.hust.edu.cn/hTFtarget),to provide searching and querying service.Users could search results with diverse methods,including query targets of given TFs or vice versa.Moreover,users could practice conjunctive queries,batch queries,and filtering query results.Secondly,we tried to explore the regulatory mechanism of TFs with these prediction results.We evaluated the association between transcription factor binding and epigenomic states,and found that 1)in many cases,the top ranked peaks of TFs are tend to bind to genome regions with Active TSS and Flanking Active TSS states,2)the peaks from multiple datasets of a same TF or members of the same TF family are tend to enriched in similar epigenomic states.Next,we analyzed the tissue-specific targets of 14 TFs among 10 different cell lines,and discussed the function of TFs according to functional enrichment results of their targets.We also applied a machine learning algorithm to impute TF co-association among different cell lines,and obtained relative importance(RI)profile of partner factors to focus-factor,which provide a quantitative perspective to estimate TF co-association.Lastly,we analyzed the role of TFs in IncRNA regulation,and the impacts of SNP in transcription factor binding.We analyzed the regulatory effect of TFs to different types of genes,and found TFs are tend to regulate protein-coding genes and IncRNA genes.We detected 9,815,083 SNPs locate on transcription factor binding sites,and 231,558 of them are located on promoter region of IncRNA genes,and these SNPs may result in the loss of function on individual transcription factor binding sites.By comparing these SNPs,which could cause TFBS losing,with IncRNA-SNPs collected in IncRNASNP database,a database curated IncRNA and SNP interaction,we found 68,597 SNPs located in lncRNASNP.As a summary,we proposed systematically predictions of human TF targets,and organized the results into hTFtarget database.Meanwhile we studied the tissue-specific regulation,co-association and IncRNA regulation of TFs,these data and results provided solid resources for gene expression studies and important clues for understanding the regulatory mechanism of TFs.更多还原