Statistical Methods for Single-Cell RNA-Seq

单细胞RNA测序（scRNA-seq）最近已经成为在单个细胞水平上研究转录组变异和调控的强大技术。传统的“批量”RNA-seq从大量细胞中汇集RNA，并测量样品中的平均表达。相比之下，scRNA-seq揭示了细胞与细胞的异质性，为理解发育、分化和疾病病因学中的生物学过程提供了关键信息。这项新技术在基础和临床研究中的应用范围不断扩大，但也因其独特的数据特征而给分析带来了挑战。其中包括：1）由于少量的起始材料和额外的样品制备程序，在存在技术伪影的情况下难以估计分子计数; 2）缺乏用于单细胞RNA计数数据的功能聚类的适当方法，其比批量RNA-seq稀疏得多; 3）缺乏定量测量和异质性的比较。我们建议通过开发一系列用于scRNA-seq数据预处理和分析的新统计方法来应对这些挑战。这包括消除RNA捕获和扩增中的技术偏差，以获得准确的分子水平计数，鉴定细胞的功能类型/亚型和可解释的特征组，解释样品和细胞之间的异质性，并鉴定差异异质性。在这个项目中开发的所有方法将作为免费的开源软件实施和发布，以造福基因组学研究社区。该方案建立的概率模型和统计框架将为其他单细胞测序实验（如单细胞ATAC-seq或BS-seq）的方法学开发奠定基础。

Single-cell RNA-sequencing (scRNA-seq) has emerged very recently as a powerful technology to investigate transcriptomic variation and regulation at the individual cell level. Traditional "bulk" RNA-seq pools RNA from a large number of cells and measures the averaged expressions in a sample. In contrast, scRNA-seq reveals cell to cell heterogeneity, providing critical information to the understanding of biological processes in development, differentiation, and disease etiologies. This new technology leads to an expansion of applications in both basic and clinical research, but also brings challenges in analysis with its unique data characteristics. These include: 1) difficulty in estimating molecule counts with the presence of technical artifacts, due to small amount of starting material and additional sample preparation procedures; 2) lack of appropriate methods for functional clustering for single cell RNA count data, which are much sparser than bulk RNA-seq; 3) lack of a quantitative measure and comparison of heterogeneity. We propose to address these challenges by developing a series of novel statistical methods for scRNA-seq data preprocessing and analyses. This includes removing technical bias in RNA capture and amplification to obtain accurate molecule level counts, identifying functional types/ subtypes of cells and interpretable feature groups, explaining heterogeneities between samples and cells, and identifying differential heterogeneity. All methods developed in this project will be implemented and released as free, open source software to benefit the genomics research community. The probability model and statistical framework established in this proposal will lay a foundation for future methodology development for other single cell sequencing experiments such as single-cell ATAC-seq or BS-seq.