集成单细胞转录组数据的选择性多聚腺苷化网络建模研究

Alternative polyadenylation (APA) is one of the essential post-transcriptional processes in eukaryotic cells, which is widespread in eukaryotes and is dynamically changing and increasingly associated with diseases. Analysis of APA dynamics in single cells contributes to cell population heterogeneity and cell typing studies. Based on our studies on polyadenylation in the past several years, first we will identify genome-wide APA sites and differentially expressed APA from 3'-end sequencing, bulk RNA-seq and single-cell RNA-seq data using various models such as bridge and link based transcriptome assembly and the generalized linear model. Next, based on multiple instance learning, shrinkage canonical correlation analysis and other methods, we will build the APA isoform network and APA gene co-expression network and fuse networks by integrating transcriptome data from different sources. We will also design network propagation strategy based on random walk to detect APA pathways and study APA specificity and conservation. Moreover, we will build single-cell APA network at the single cell level and adopt network propagation strategy to identify cell-specific APA and cell types by network embedding. Finally, we will release our data as an open-access database and create a web-based server for the analysis and visualization of APA in various species and conditions. This project focuses on network modeling at APA and single-cell levels to elucidate the functional diversity of APA from a systems biology perspective. This project intends to identify APA genes and sites associated with specific biological functions, diseases or cell types, and elucidate specificity, conservation and heterogeneity of APA in different tissues and cell types. This study will provide important information for APA-related development or diseases and will have broad application prospects in genomics and biomedicine.

选择性多聚腺苷化(APA)是重要转录后调控过程，在真核生物中广泛存在、动态变化且与疾病的关联日益显现。单细胞APA动态的分析有助于细胞群异质性及细胞分型研究。基于多年研究基础，本项目拟基于转录组装配、广义线性模型等方法从3′末端、传统及单细胞转录组数据识别APA位点和APA使用差异；基于多示例学习、收缩典型相关分析等方法构建APA转录本网络和基因共表达网络并实现网络融合；基于随机游走策略实现APA通路检测、特异性和保守性研究；构建单细胞APA网络，基于网络传播识别细胞特异APA及嵌合多维APA网络实现细胞分型；最终建立高效便捷的生物信息平台。此研究在APA及单细胞水平进行网络建模研究，从系统生物学的角度揭示APA的功能多样性，能识别与特定生物功能、疾病、发育或细胞类型有关的APA基因和位点，阐释APA在不同组织的特异性和保守性及单细胞水平的异质性，将在基因组学和生物医学中有广泛应用前景。

本项目从APA转录本及单细胞水平，探索细胞类型特异的APA位点调控机制。此研究能从系统生物学的角度揭示APA的功能多样性，识别与特定生物功能、通路、疾病或细胞类型有关的APA基因和位点，能为APA相关的发育或疾病提供重要信息，将在基因组学和生物医学中有广泛应用前景。.在论著方面，目前已资助发表了负责人为第一或通讯（含共同）的SCI论文13篇（第一标注10篇），发表于生物信息学领域的顶级和优秀刊物，如Brief in Bioinform、Genomics, Proteomics & Bioinformatics、Bioinformatics、PLoS comput bio等。.在人才培养方面，指导了2名博士生和4名硕士生完成毕业答辩并拿到学位。.在工具平台方面，已经顺利开发且公开发布了用于单细胞转录组和APA研究的工具及平台9个（https://github.com/BMILAB/），包括针对最新的单细胞RNA-seq，设计了基于全基因组分段及poly(A)锚定算法的scAPAtrap工具；针对单细胞数据高稀疏高噪声难点及小样本问题，提出了基于加权距离和SMOTE的缺失填充方法scHinter及基于转录调控网络的ADImpute方法；实现了不同3′末端文库技术及不同物种通用的普适性APA位点识别及注释工具movAPA；开发了当前最全的植物APA位点数据库PlantAPAdb（http://www.bmibig.cn/ plantAPAdb/），可用于挖掘不同组织发育、品系培育的重要功能基因。这些工具被OMICtools等生物信息常用平台推荐，被同行广泛引用。例如，movAPA和scAPAtrap工具及测评综述结果被Nat Commun、Nucleic Acids Res、Brief in Bioinform等顶级刊物上的论文作为重要依据或代表性工作引用。PlantAPAdb工作一经发布，数据便被Nat Plants上发表的最新研究用于分析转录后调控过程中AS和APA的关系。该工作也被顶级刊物Genome Biol、植物学顶刊Plant Physiol等发表的综述或研究作为APA研究的代表性资源重点介绍和引用。这些结果及工具平台将为有关单细胞转录组和APA相关的分析提供丰富的资源及有助于生物实验筛选高质量的候选基因或位点，促进APA机制的研究。