Statistical methods for analysis of single-cell variability

 描述(申请人提供)：健康和患病的组织都是由多种细胞类型组成的，这些细胞类型的相互作用支撑了它们的功能。即使在给定的细胞类型中，细胞 由于外界影响、该细胞的历史或统计事件，它们的转录状态不同。这种异质性的影响在癌症治疗的背景下尤其显著，在癌症治疗中，表型不同的亚克隆群体的存在加剧了复发和对治疗的抵抗。作为与凯瑟琳·吴(DFCI)实验室持续合作的一部分，我们正在应用单细胞基因组分析来研究白血病细胞中的亚群动态。检查慢性淋巴细胞白血病(CLL)患者的样本，我们发现显著的转录和表观遗传异质性(Landau，癌细胞在PRESS)，并旨在表征与治疗耐药相关的转录亚群，并建立它们与更好研究的基因亚克隆的关系。而单细胞分析提供了 尽管这是解剖异质组织的直接手段，但由于缺乏敏感的统计工具进行分析，它们的应用目前受到限制。在这里，我们提出了新的统计方法的开发和应用，用于从常规的和空间分辨的单细胞转录组测量中识别和表征生物上不同的细胞亚组。在我们对单细胞转录组数据进行统计建模的方法(Kharchenko，自然方法2014)的基础上，我们建议：1)使用基于模型的敏感因子分析来捕捉转录变异性的结构；2)实现一个框架来探索所有统计上的 这些研究包括：1)分析细胞群体内异质性的重要方面，从而能够重点分析生物相关的异质性；3)开发综合方法，在单细胞水平上对转录和遗传肿瘤亚群进行比对；4)将误差模型与统计小波分析和马尔可夫随机场方法相结合，从空间分辨的RNA-SEQ数据中识别异质性的空间模式，并对肿瘤和正常组织中的组织微体系结构进行建模。鉴于临床上需要进一步了解肿瘤内的异质性及其对治疗反应和耐药性的影响，我们将这些方法的应用重点放在分析白血病患者肿瘤中的亚克隆群体上。使用单细胞基因组分析，我们将检查从正在接受治疗的CLL患者的连续时间点收集的样本，作为吴实验室单独工作的一部分，已经收集了匹配的批量DNA和RNA测序数据。应用所提出的方法，我们将1)表征亚克隆群体内和亚克隆群体之间的转录异质性，2)寻找与亚克隆扩增率和外周血中针对B细胞受体途径的治疗耐药性有关的功能特征，3)应用空间分辨RNA-SEQ方法研究耐药白血病细胞的聚焦淋巴储存库。我们希望这些研究能为与治疗相关的细胞特性提供有价值的见解。 阻力，并产生广泛适用的计算工具。

 DESCRIPTION (provided by applicant): Both healthy and diseased tissues are composed of multiple cell types whose interplay underpins their functions. Even within a given cell type, cells differ in their transcriptional state due to external influences, the history of that cell, or stocastic events. The impact of such heterogeneity is particularly notable in the context of cancer therapy, where presence of phenotypically distinct subclonal populations fuels relapse and resistance to treatment. As part of an ongoing collaboration with the laboratory of Catherine Wu (DFCI), we are applying single-cell genomic assays to investigate subpopulation dynamics in leukemia cells. Examining samples from patients with chronic lymphocytic leukemia (CLL), we find notable transcriptional and epigenetic heterogeneity (Landau, Cancer Cell in press), and are aiming to characterize transcriptional subpopulations associated with therapeutic resistance and establish their relationship to better-studied genetic subclones. While single-cell assays provide direct means to dissect heterogeneous tissues, their application is currently limited by the lack of sensitive statistical tools for their analysis. Here we propose the development and application of novel statistical methods for the identification and characterization of biologically distinct subsets of cells from regular and spatially-resolved single-cell transcriptome measurements. Building on our approach for statistical modeling of single-cell transcriptome data (Kharchenko, Nature Methods 2014) we propose to: 1) use sensitive model-based factor analysis to capture the structure of transcriptional variability; 2) implement a framework to explore all statistically significant aspects of heterogeneity within cell population, enabling a focused analysis of biologically relevant heterogeneity; 3) develop integrative approach to align transcriptional and genetic tumor subpopulations on a single-cell level; 4) combine error models with statistical wavelet analysis and Markov Random Field methods to identify spatial patterns of heterogeneity from spatially- resolved RNA-seq data, and model tissue microarchitecture in tumor and normal tissue. Given the clear clinical need to advance our understanding of intra-tumor heterogeneity in cancer and its impact on therapy response and resistance, we focus the application of these methods on the analysis of subclonal populations in tumors of leukemia patients. Using single-cell genomic assays, we will examine samples collected at serial time points from CLL patients undergoing therapy, for which matched bulk DNA and RNA sequencing data have been collected as part of a separate effort by the Wu lab. Applying the proposed methods, we will 1) characterize transcriptional heterogeneity within and between the subclonal populations, 2) search for functional features linked to subclonal expansion rates and resistance to therapies targeting B-cell receptor pathway in peripheral blood; 3) apply spatially-resolved RNA-seq methods to investigate focused lymph node reservoirs of drug-resistant leukemic cells. We expect these studies to provide valuable insights into cell characteristics associated with therapy resistance, and yield widely applicable computational tools.