Collaborative Research: Advanced statistical methods for single cell RNA sequencing studies

单细胞RNA测序已经成为基因组学中的一种强有力的工具，并在广泛的 各种应用，为许多基本的生物学问题提供了前所未有的洞察力 以前很难解决。然而，分析scRNAseq数据面临着重要的统计和 需要开发新的计算和统计方法的计算挑战。 主要挑战包括：(1)缺乏可靠的统计方法来控制隐藏的混杂 在一系列环境中的影响；(2)缺乏准确的细胞亚群聚类方法， 为scRNAseq研究量身定做；以及(3)用scRNAseq识别功能性遗传变异的难度 在将scRNAseq与其他基因研究整合方面的单独和困难包括全基因组关联 学习。我们提出的方法将应对这些挑战，并在以下方面具有创新性：(1) 我们控制隐藏混杂效应的方法在两类现有的统计之间架起了桥梁 消除混杂影响的方法，因此有望在一系列 (2)我们的细胞亚群聚类方法从低维数据中提取聚类信息 表示scRNAseq数据，因此即使在以下情况下也有望产生准确结果 原始的高维基因表达矩阵有噪声；以及(3)我们的等位基因识别方法 仅使用scRNAseq数据的特定/有偏见的表达代表了第一次这样的尝试和我们的方法 将scRNAseq与GWAS集成也是此类尝试的第一次。我们建议的所有方法都是 为scRNAseq数据量身定做，并将处理scRNAseq数据的复杂性和独特性， 包括但不限于低覆盖率、计数性质和辍学事件。我们将开发、分发、 并支持用户友好的开源软件实现我们的方法，以造福于基因组学和 统计界。本文开发的统计方法将为开发类似的方法铺平道路 其他测序研究，包括亚硫酸氢盐测序和ATAC-SEQ研究。建议的方法 对于理解组织成分的异质性和肿瘤的遗传结构是必不可少的 复杂的特征和疾病--两者都是对人类健康至关重要的问题。

Single cell RNA sequencing has emerged as a powerful tool in genomics and has been used in a wide variety of applications, providing unprecedented insights into many basic biological questions that are previously difficult to address. However, analyzing scRNAseq data face important statistical and computational challenges that require the development of new computational and statistical methods. Key challenges include: (1) lack of robust statistical methods that can control for hidden confounding effects in a range of settings; (2) lack of accurate cell subpopulation clustering methods that are tailored to scRNAseq studies; and (3) difficulty in identifying functional genetic variations with scRNAseq alone and difficulty in integrating scRNAseq with other genetic studies include genome-wide association studies. Our proposed methods will address these challenges and are innovative in the following aspects: (1) our method for controlling for hidden confounding effects bridges between two existing classes of statistical methods for removing confounding effects and is thus expected to perform robustly across a range of scenarios; (2) our method for clustering cell subpopulations extracts clustering information from a lowdimensional representation of scRNAseq data and is thus expected to produce accurate results even when the original high-dimensional gene expression matrix is noisy; and (3) our method for identifying allele specific/biased expression using scRNAseq data alone represents the first such attempt and our method for integrating scRNAseq with GWASs also represents the first such attempt. All our proposed methods are tailored to scRNAseq data and will cope with the complexities and unique features of scRNAseq data, including, but not limited to, low-coverage, count nature, and drop-out events. We will develop, distribute, and support user-friendly open-source software implementing our methods to benefit the genomics and statistics community. The statistical methods developed here will pave ways for developing similar methods to other sequencing studies including bisulfite sequencing and ATAC-seq studies. The proposed methods are essential for understanding the heterogeneity of tissue compositions and the genetic architecture of complex traits and diseases - both are questions of central importance to human health.