Computational Inference of Regulatory Network Dynamics on Cell Lineages

控制哪些基因在何时表达的调控网络，在维持和 不同细胞状态的转变。在哺乳动物系统中，这样的网络是通过复杂的相互作用建立的 在数千种调控蛋白中，如转录因子、染色质重构体和信号蛋白， 组蛋白翻译后修饰和基因组的三维组织。因此， 识别基因组规模的调控网络及其变化仍然是一项计算和 实验挑战，特别是对于稀有和新的细胞类型。通过财团最近的努力 项目我们现在拥有丰富的数据集，在模型单元中测量调节机械的多个组件 台词。这些数据使我们能够为这些细胞系建立一个更完整的调控网络。我们可以使用 此信息用于识别新小区类型中的网络，其中仅测量 调节机制是可能的(例如转录组)？我们能否利用更完整的监管网络 预测新的细胞类型，并预测网络扰动对细胞状态的影响？要解决这些问题 问题，在这项建议中，我们将制定创新的网络重建方法，以确定 通过利用它们与研究充分的细胞的关系，在新的和稀有的细胞类型中调节网络 类型，以及彼此之间的关系。我们的方法将使用非静态图形模型的框架来 代表特定于细胞类型的监管网络，并将使用多任务学习来整合共享 谱系中细胞类型之间的信息。目标1中的方法将推断模块化基因调控网络 每种细胞类型，并进一步完善现有的不完整或不确定的谱系结构。目标2中的方法 将确定染色质状态和转录因子之间的细胞类型特定的直接依赖关系以及如何 它们通过近距离和长期调节来影响靶基因的表达。我们的方法将应用于 两个细胞命运指定问题：细胞重编程和多细胞谱系向前分化。在……里面 蜂窝重编程、调节器和阻碍重编程效率的子网络将被预测和 使用遗传扰动实验进行了测试。在向前分化方面，监管网络的变化推动了 将确定和测试替代血统。该项目的成功完成将提供两个广泛的 适用的软件工具，使研究人员能够(I)准确识别监管网络及其 复杂细胞谱系中不同细胞状态之间的变化，(Ii)检查多个水平之间的相互作用 及其对细胞类型特定基因表达的影响，以及(Iii)有效地识别最 上游调控基因和子网络改变细胞状态。来自该项目的软件工具将是 提供并将广泛适用于发展中的各种类型的动态生物过程 和疾病。

Regulatory networks that control which genes are expressed when, are critical players in the maintenance and transitions of different cell states. In mammalian systems such networks are established by a complex interplay of thousands of regulatory proteins such as transcription factors, chromatin remodelers and signaling proteins, histone post-translational modifications and three-dimensional organization of the genome. Hence, the identification of genome-scale regulatory networks and their changes remains a computational and experimental challenge, especially for rare and novel cell types. Through recent efforts of consortia projects we now have rich datasets measuring multiple components of the regulation machinery in model cell lines. These data enable the creation of a more complete regulatory network for these cell lines. Can we use this information to identify networks in new cell types where measuring only a few components of the regulation machinery is possible (e.g. the transcriptome)? Can we leverage more complete regulatory networks to predict new cell types, and to predict the effect of network perturbations to cellular state? To tackle these questions, in this proposal we will develop innovative network reconstruction methods to identify regulatory networks in novel and rare cell types by leveraging their relationships to well-studied cell types, as well as to each other. Our methods will use the framework of non-stationary graphical models to represent cell type-specific regulatory networks and will use multi-task learning to incorporate shared information between cell types in a lineage. Methods in Aim 1 will infer modular gene regulatory networks for each cell type and additionally refine an existing incomplete or uncertain lineage structure. Methods in Aim 2 will identify cell type-specific directed dependencies among chromatin state and transcription factors and how they impact target gene expression through proximal and long-range regulation. Our methods will be applied to two cell-fate specification problems: cellular reprogramming and multi-cell lineage forward differentiation. In cellular reprogramming, regulators and subnetworks hindering reprogramming efficiency will be predicted and tested using genetic perturbation experiments. In forward differentiation, regulatory network changes that drive alternate lineages will be identified and tested. Successful completion of this project will provide two broadly applicable software tools that will enable researchers to (i) accurately identify regulatory networks and their changes between different cell states in complex cell lineages, (ii) examine interactions among multiple levels of regulation and their impact on cell type-specific gene expression, and (iii) efficiently identify the most upstream regulatory genes and subnetworks that change cellular states. Software tools from this project will be made available and will be broadly applicable to diverse types of dynamic biological processes in development and disease.