Studying temporal dynamics and regulatory mechanisms of single cells with a unified framework and multi-omics data

单细胞基因组学技术使研究人员能够研究单细胞之间的差异。按顺序 为了了解整个生物体中每个细胞的功能，需要在以下背景下研究细胞 时间和空间。研究人员希望全面了解每个单细胞，包括其 当前的细胞状态和预测的未来状态，以及它在时间期间如何与邻近细胞相互作用 动态。所以在获得一定的单细胞知识后，下一步就是从“零件”到“零件”。 “整体”。该提案讨论了可以为时间动力学研究带来的进展 以及细胞之间的空间相互作用。该提案的主题是“整合”。 现有的单细胞数据整合方法主要集中在两种场景：1）整合相同类型的数据（例如 来自多个批次的RNA-Seq数据； 2）整合对同一细胞执行的多种类型的数据，即 也称为多模态数据或多组学数据。该提案强调了以下概念和方法： 整合多组学数据以了解细胞时间动态和调控机制，同时 考虑到数据模式之间的依赖性，这在当前方法中很少见。一个新的 “问题集成”的概念，其中相关的计算公式可以连接到 此处还提供了细胞某个方面的一致且更一般的图片。 为了研究细胞的不同方面，已经制定了不同的计算问题，例如， 细胞聚类、细胞轨迹推断、基因调控网络（GRN）推断等。 统一或连接相关的计算问题，以便统一的框架可以涉及或 提出了输出先前在多个单独计算问题中使用的信息。在 特别是，涉及相关计算的细胞时间动力学分析的统一框架 任务，提出。 到目前为止，多组学集成方法通常将集成数据部署到针对细胞类型的细胞聚类 识别。很少有数据集成方法是为连续群体的时间分析而设计的， 或者学习 GRN 等生物机制。所以这里提出的另一个方向是推断 具有基因表达 (scRNA-seq) 和染色质可及性 (scATAC-seq) 数据的细胞；与推断的 轨迹，可以研究染色质可及性对基因表达的影响，并且可以 重建时考虑了这种影响。 实际上，基因的表达水平由多种因素决定：转录因子 (TF)、染色质 可访问性以及小区通过小区间交互从其他相邻小区接收的信号。因此， 关于细胞需要考虑的另一个重要维度是细胞的空间位置。建议 ß重建一个广义的 GRN，用于模拟细胞间的调节相互作用。

Single cell genomics technologies have allowed researchers to study differences between single cells. In order to understand how every cell functions in the whole living organism, cells need to be studied in the context of both time and space. Researchers would like to learn a comprehensive picture of each single cell, including its current cell state and predicted future state, and how it interacts with neighboring cells during the temporal dynamics. So the step after gaining a certain amount of knowledge of single cells is to go from “parts” to “whole”. This proposal discusses advances that can be brought to the study of both temporal dynamics and spatial interactions between cells. The theme of this proposal is “integration”. Existing integrative methods for single cell data focus on two scenarios: 1) integrate the same type of data (eg. RNA-Seq data) from multiple batches; 2) integrate multiple types of data performed on the same cells, which is also called multi-modality data or multi-omics data. This proposal highlights concepts and methods to integrate multi-omics data to understand cell temporal dynamics and regulatory mechanisms, while taking into account dependency between data modalities, which is rare in current methods. A new concept of “problem integration”, where related computational formulations can be connected to provide a consistent and more general picture of a certain aspect of cells, is also presented here. In order to study different aspects of a cell, different computational problems have been formulated, eg., clustering of cells, inference of cell trajectories, inference of gene regulatory networks (GRNs), etc. The idea of unifying or connecting related computational problems, such that a unified framework can involve or output the information that is previously used in multiple individual computational problems, is proposed. In particular, a unified framework for cell temporal dynamics analysis involving related computational tasks, is presented. So far the multi-omics integration methods often deploy the integrated data to cluster cells for cell type identification. Few methods on data integration are designed for temporal analysis with continuous populations, or to learn biological mechanisms like GRNs. So another direction proposed here is to infer the trajectory of cells with both gene-expression (scRNA-seq) and chromatin accessibility (scATAC-seq) data; with the inferred trajectory, the effect of chromatin accessibility on gene-expression can be studied, and GRNs can be reconstructed while taking into account this effect. In reality, a gene’s expression level is determined by multiple factors: its transcription factor (TF), its chromatin accessibility and the signal a cell receives from other neighboring cells through cell-cell interaction. Therefore, another important dimension to consider about the cells is the spatial location of cells. It is proposed to ßreconstruct a generalized GRN which models inter-cell regulatory interactions.