Computational methods to predict gene regulatory network dynamics and cell state transitions

项目摘要 这项研究计划的目标是提供工具，以发现控制 细胞命运的决定。由细胞状态转换驱动的细胞命运决定是基本的细胞过程的基础 发展到细胞重新编程。有机会利用可公开获得的基因组数据 发展细胞状态转变动力学的预测性计算模型。建议的方法将提供 在给定特定细胞的情况下，洞察细胞命运决策及其转录调控方式的手段 命运决策点和合适的数据。这种决策点的例子包括对表皮的控制 再生，或在造血过程中维持髓系细胞命运的平衡。为了架设桥梁 基因组学和细胞动力学、统计和计算建模之间的差距必须是 克服困难。这项研究计划的基础是两个关键挑战：1)开发统计方法以 推断调控网络，同时考虑到单个细胞之间的可变性水平，以及2)发展 耦合细胞内基因调控动力学和细胞间通讯的计算模型 细胞。为了解决第一个挑战，我们将开发机器学习模型来预测基因表达 来自时间序列数据的动力学。这些模型将能够根据它们的时间模式对基因进行分类，并且 这一结果将为基因网络推理提供依据。然后我们将开发用于网络推理的方法 集成多模式数据(单细胞RNA和ATAC测序)以及细胞-细胞信号信息以 学习控制特定小区状态转换的网络。为了应对第二个挑战，我们将制定 基于微分方程式的基因调控网络动力学的多尺度模型与细胞- 外部信令动态。这将使我们能够同时捕捉分子和细胞动力学。 分辨率，从而确定哪些参数对系统施加关键控制。我们将使用贝叶斯 用于参数推理以使模型与数据相匹配并执行模型选择的方法，适用于 多尺度模型推理所需的。将通过将模型应用于特定项目来对其进行严格评估 系统，包括细胞分化(例如，在造血过程中决定髓系的命运)和发育(例如， 肾单位祖细胞命运的决定)。在这些器官系统中的每一个中，模型预测都将得到检验 通过合作进行实验。在迭代测试之后，将制定开源的、经过验证的方法 广泛用于研究细胞命运决策的动态过程。

Project Summary The goal of this research program is to provide tools for the discovery of transcriptional networks that control cell fate decisions. Cell fate decisions driven by cell state transitions underlie essential cell processes from development to cellular reprogramming. There is an opportunity to make use of publicly available genomic data to develop predictive computational models of cell state transition dynamics. The methods proposed will offer means to gain insight into cell fate decision-making and how it is transcriptionally regulated, given specific cell fate decision points and suitable data. Examples of such decision points include control of epidermal regeneration, or the maintenance of balance among myeloid cell fates during hematopoiesis. In order to bridge the gap between genomics and cell dynamics, statistical and computational modeling challenges must be overcome. Two key challenges form the basis of this research program: 1) developing statistical methods to infer regulatory networks while accounting for the levels of variability between single cells, and 2) developing computational models to couple gene regulatory dynamics within cells and cell-cell communication between cells. To address the first challenge, we will develop machine learning models to predict gene expression dynamics from time-series data. These models will be able to classify genes by their temporal patterns, and the results will inform gene network inference. We will then develop methods for network inference that integrate muti-modal data (single-cell RNA and ATAC sequencing) as well as cell-cell signaling information to learn networks that control specific cell state transitions. To address the second challenge, we will develop differential equation-based multiscale models of the gene regulatory network dynamics coupled with the cell- external signaling dynamics. This will allow us to capture both molecular and cellular dynamics in high resolution, and thus identify which parameters exert key control over the system. We will use Bayesian methods for parameter inference to fit models to data and perform model selection, adapting methods where needed for multiscale model inference. Models will be rigorously evaluated through their application to specific systems, including cell differentiation (e.g. myeloid fate decisions during hematopoiesis) and development (e.g. nephron progenitor cell fate decisions). In each of these organ systems, models predictions will be tested experimentally via collaborations. Following iterative testing, open-source, validated methods will be made widely available for the study of the dynamic processes of cell fate decision-making.