The transcriptome-wide impact of biological perturbations

摘要 计算生物学的一个重要目标是利用来自高通量功能分析的数据来推断 基因变异的生物学后果。这一目标通常通过配对RNA测序来实现 和差异表达分析。大多数差异表达方法试图识别一小部分 受基因扰动影响的基因和基因组。然而，一些基因，如染色质 监管机构，可能会影响整个转录组的数千个基因。这些分散的效果不会被捕获 通过现有的方法。我们将通过以下方式解决差异表达领域中的方法论差距 开发一种新的统计工具，并将把这一工具应用于规范和疾病方面。在……里面 目标1，我们提出了翻译全影响模型(TIM)，这是一种基于参数似然的估计 扰动对转录组的整体影响。TIM建立在现有差异表达方法的基础上， 但估计的是差异表达效应的分布参数，而不是每个基因的个体 效果大小。该模型还被扩展到估计基因集丰富度和差异之间的相关性 表情签名。在目标2中，我们的目标是将TIM应用于最近的扰动序列数据集，该数据集表达了所有扰动 以大规模平行的方式在体外表达基因，使我们能够识别哪些基因和基因集诱导 人类慢性粒细胞白血病细胞系被击倒时最大的转录变化。我们还将 使用TIM识别对转录组有类似影响的基因模块，并使用这些模块来 诠释基因功能。在目标3中，我们将TIM应用于35个神经发育的体内扰动-序列数据集 发育中的小鼠新皮质中的紊乱基因。通过这个目标，我们将对神经发育障碍进行分层 基因按转录组影响的程度，测试神经发育障碍- 相关的基因表达调控因子对大脑中的转录组起着高度分散的作用。如果为真，则这 这一发现将引发一个耐人寻味的问题，即微小的分散表达效应是否会致病。 为研究神经发育障碍以及许多其他疾病开辟了新的途径 与表达调节因子(如癌症)相关。我们还将使用TIM对神经发育进行分类 通过对紊乱基因转录效应的相似性分析，找出具有推测收敛机制的基因。 总的来说，我们的模型将允许从差异表达实验中提取概念上的新见解， 具有对任何感兴趣的生物扰动的适用性。

Abstract An important goal in computational biology is to leverage data from high-throughput functional assays to infer the biological consequences of genetic variation. This goal is frequently approached by pairing RNA sequencing and differential expression analysis. Most differential expression methods seek to identify a small number of genes and gene-sets that are affected by a genetic perturbation. However, some genes, such as chromatin regulators, may impact thousands of genes across the transcriptome. These dispersed effects are not captured by existing methods. We will address this methodological gap in the differential expression field by developing a novel statistical tool, and will apply this tool to both normative and disease contexts. In Aim 1, we propose the Transcriptome-wide Impact Model (TIM), a parametric likelihood-based estimator of the overall effect that a perturbation has on the transcriptome. TIM builds on existing differential expression methods, but estimates parameters of the distribution of differential expression effects, rather than individual per-gene effect sizes. This model is also extended to estimate gene-set enrichments and correlation between differential expression signatures. In Aim 2, we aim to apply TIM to a recent Perturb-Seq dataset that perturbs all expressed genes in vitro in a massively parallel manner, enabling us to identify which genes and gene-sets induce the greatest transcriptomic change in human chronic myeloid leukemia cell lines when knocked down. We will also use TIM to identify modules of genes that have similar impact on the transcriptome, and use these modules to annotate genic function. In Aim 3, we will apply TIM to an in vivo Perturb-Seq dataset of 35 neurodevelopmental disorder genes in developing mouse neocortex. Through this Aim, we will stratify neurodevelopmental disorder genes by degree of transcriptome-wide impact, testing the hypothesis that neurodevelopmental-disorder- associated gene expression regulators exert highly dispersed effects on the transcriptome in brain. If true, this finding would raise the intriguing question of whether small, dispersed expression effects can be pathogenic, opening novel avenues for research into neurodevelopmental disorders, as well as many other diseases that are associated with expression regulators (e.g. cancer). We will additionally use TIM to cluster neurodevelopmental disorder genes by similarity of transcriptomic effects, to identify genes with putatively convergent mechanism. Broadly, our model will allow conceptually novel insight to be extracted from differential expression experiments, with applicability to any biological perturbation of interest.