Computational tools for estimating cell-type-specific effects in bulk RNA-seq and spatial transcriptomics data, using reference single-cell RNA-seq datasets

项目摘要/摘要 RNA-SEQ是研究分子生物学的有力工具。然而，在没有细胞分类(或相关技术)的情况下， 应用于组织样本的常规RNA-SEQ不能确定潜在细胞类型的基因表达。 这是有问题的，因为在组织水平上观察到的差异基因表达不一定反映出来 在下面的细胞类型中，这模糊了生物学上的洞察力。例如，Schmiedel等人。最近应用的RNA- 对106名个体的13种纯化血细胞类型进行了序列分析，揭示了性别特异性的分子基础 免疫反应的差异。然而，当他们只将rna-seq应用于全血时，这一点就被掩盖了。 单细胞rna-seq是更广泛地探索细胞类型特定效应的明显候选者。然而，对于大多数人来说， 由于特殊的解离方案，单细胞rna-seq被限制在小样本大小的组织中。 和成本。因此，只有散装组织RNA-seq数据可用于大样本。至关重要的是，其中大部分 大量数据与信息量巨大的临床表型数据和附加组学数据配对。这些 数据集包括大型NIH倡议，如GTEx，TCGA，以及我们所有人，它们收集了遗传学数据， 疾病状况、结局、药物治疗、种族、性别等。关键的差距是我们不能 目前研究细胞类型水平的基因表达与这些表型之间的关系。 为了克服这一限制，我们将开发用于估计细胞类型特定差异的计算工具 批量rna-seq数据的表达式，当可从 同样的组织类型。这将使我们能够研究驱动人类基因表达的细胞类型特异性差异 表型和疾病，解锁了与表型数据配对的数以万计的批量RNA-SEQ样本。 这个研究计划的基础是我们之前的一项研究，在那里我们开发了一种方法来恢复细胞类型- 大量乳腺肿瘤rna-seq数据中遗传遗传变异对基因表达的特殊影响。这种方法 这让我们发现了一种新的乳腺癌风险基因--这一基因被传统方法掩盖了。 在这里，我们假设类似的数学框架可以适用于恢复任何细胞类型特定的效应 从大块组织RNA-seq.因此，我们可以开发特定的工具来执行多个常用分析 通过利用匹配的单细胞数据从块状组织RNA-seq中获得特定细胞类型的分辨率，包括 差异表达、相关性和基因集富集性分析。 最后，新的空间转录组学技术正在出现，它使空间分辨的基因表达能够 直接在组织切片中测量。这些平台在条形码的~100μm中原位量化基因表达 斑点。每个斑点捕获一小群细胞--类似于微型的大块组织rna-seq实验。 因此，可以使用相同的抽象数学框架来识别诸如特定于细胞类型的影响 基因表达的空间变异。这些数据的计算工具正在迅速发展；因此，该奖项将 也使我们能够开发出满足这些新基因表达平台不断变化的需求的方法。

PROJECT SUMMARY / ABSTRACT RNA-seq is a powerful tool for studying molecular biology. However, without cell sorting (or related techniques), conventional RNA-seq applied to tissue samples cannot determine gene expression in underlying cell-types. This is problematic because differential gene expression observed at the tissue level is not necessarily reflected in underling cell-types, which obscures biological insight. For example, Schmiedel et al. recently applied RNA- seq to 13 purified blood cell-types from 106 individuals1, which uncovered the molecular basis of sex-specific differences in immune response. However, this was obscured when they applied RNA-seq to only whole-blood. Single-cell RNA-seq is the obvious candidate to probe cell-type-specific effects more broadly. However, for most tissues, single-cell RNA-seq has been restricted to small sample sizes, due to specialized dissociation protocols and cost. Thus, only bulk-tissue RNA-seq data are available for large sample sizes. Crucially, much of these bulk data are paired to enormous stores of informative clinical phenotypic data and additional -omics data. These datasets include large NIH initiatives such as GTEx, TCGA, and All of Us, which have collected data on genetics, disease status, outcome, drug treatments, ethnicity, sex, and much more. The critical gap is that we cannot currently study the relationship between cell-type level gene expression and any of these phenotypes. To overcome this limitation, we will develop computational tools for estimating cell-type-specific differential expression from bulk RNA-seq data, when a small reference single-cell RNA-seq dataset is available from the same tissue-type. This will allow us to study the cell-type-specific differences in expression that drive human phenotypes and diseases, unlocking the tens-of-thousands of bulk RNA-seq samples paired to phenotypic data. The basis for this research program is a previous study where we developed a method to recover the cell-type- specific effects of inherited genetic variation on gene expression in bulk breast-tumor RNA-seq data. This method allowed us to discover a novel breast cancer risk gene—which was obscured using conventional methods. Here, we posit that a similar mathematical framework can be adapted to recover any cell-type-specific effect from bulk-tissue RNA-seq. Hence, we can develop specific tools to perform multiple commonly applied analyses at cell-type-specific resolution from bulk-tissue RNA-seq by leveraging matched single-cell data, including differential expression, correlative and gene set enrichment analysis. Finally, new spatial transcriptomics technologies are emerging that enable spatially resolved gene expression to be measured directly in tissue sections. These platforms quantify gene expression in situ in ~100μm barcoded spots. Each spot captures a small cluster of cells—akin to a miniaturized bulk-tissue RNA-seq experiment. Hence, the same abstract mathematical framework can be used to identify effects such as cell-type-specific spatial variation in gene expression. Computational tools for these data are evolving quickly; thus, this award will also allow us to develop methods that meet the changing needs of these new gene expression platforms.