ABI Innovation: doseR: a novel framework for dosage compensation and global expression analysis

本研究旨在开发并广泛提供一种新的统计方法，用于分析染色体拷贝数差异导致的基因表达模式，特别是性染色体上出现的差异（例如XX女性与XY男性）。通常，染色体拷贝数的差异导致这些染色体的基因剂量的差异，导致对基因表达的相应影响，称为“剂量效应”。然而，在许多生物体中，特殊的“剂量补偿”机制已经发展到减轻这些剂量效应所产生的差异，在染色体拷贝，如男性和女性之间的性染色体。有越来越多的兴趣了解哪些生物体采用剂量补偿机制，以及为什么。基因表达的全基因组测定的最新进展极大地扩展了可以测定剂量补偿的分类群，但是用于这些数据的分析方法是变化的，不一致地应用，并且通常没有充分利用数据中可用的信息。从这个项目中开发的方法和软件称为“doseR”，将为这些问题中的每一个提供一个连贯和全面的解决方案。因此，“doseR”项目填补了目前存在于剂量补偿基因组研究分析方法中的一个主要空白。此外，除了剂量补偿分析之外，这种方法可以推广到识别在不同生物条件下出现的基因组之间的基因表达的广泛变化。因此，doseR的开发和部署将弥合生物直觉和生物信息学推理之间的差距，不仅用于剂量补偿，而且用于许多尚未预见的调查路线。该项目还为本科生创造了几个培训机会，包括密集的生物信息学培训研讨会和直接参与研究活动。该研究旨在开发并广泛提供一种新的线性建模统计方法，用于使用全基因组RNA-seq表达数据分析性染色体剂量补偿。鉴于数据的性质和所需的一套推论，目前用于这种分析的统计方法远非理想。目前，生物学重复被平均为每个基因的单个测量，并被高度标准化。然后，使用绝对表达评价基因表达对性染色体相对于常染色体的特定影响，同时使用表达比率评估基因剂量效应，在两种情况下均使用非参数统计检验。一种更具有统计学稳健性的方法是采用基因表达的线性混合效应模型。这提供了一个统一的统计框架，以评估染色体特异性和剂量对基因表达的影响的幅度和意义。此外，它直接应用于每个基因的测序读段计数，并将缩放因子如测序深度和转录本长度并入描述数据的模型中，如在差异表达的大多数分析中所做的那样。因此，避免了广泛的标准化，并且易于将统计重复纳入分析中。这些方法将在一个名为“doseR”的新软件包中实施，该软件包用R统计编程语言编写，并作为生物信息学软件工具的Bioconductor套件的一部分分发。新的统计模型及其软件实现相对于先前的评估剂量补偿的方法的性能将通过RNA测序数据的广泛模拟进行评估。还将检查和评估与剂量补偿相关的特定经验数据集的应用。虽然软件开发的直接动机是剂量补偿分析，但所提出的方法可用于任何需要检测多个特定基因子集表达方向变化的分析场景。因此，它提供了一个工具，在系统生物学研究中具有广泛的实用性。该项目的状态和结果可在https://walterslab.github.io/doseR/上找到。

This research aims to develop and make broadly available a novel statistical methodology for analyzing patterns of gene expression that result from differences in chromosome copy number, in particular as arise on the sex-chromosomes (e.g. XX females versus XY males). Often a difference in chromosome copy number, which causes a difference in gene dose for those chromosomes, results in a corresponding effect on gene expression, termed a "dosage effect". However, in many organisms, special "dosage compensation" mechanisms have evolved to mitigate these dosage effects arising from differences in chromosome copies, such as between males and females on the sex-chromosomes. There is increasing interest in understanding which organisms employ dosage compensation mechanisms, and why. Recent advances in genome-wide assays of gene expression have greatly expanded which taxa can be assayed for dosage compensation, but analytical approaches for such data are varied, inconsistently applied, and typically do not make full use of the information available in the data. The methods and software developed from this project, called "doseR", will provide a cohesive and comprehensive solution to each of these issues. As such, the "doseR" project fills a major gap that currently exists in analytical methods for genomic investigations of dosage compensation. Furthermore, beyond dosage compensation analysis, this methodology can be generalized to identify broad shifts in gene expression between groups of genes that arise under different biological conditions. Thus, development and deployment of doseR will bridge the gap between biological intuition and bioinformatic inference, not only for dosage compensation, but also for many still as yet unforeseen lines of inquiry. This project also creates several training opportunities for undergraduate students, including intensive bioinformatic training workshops and direct participation in research activities.This research aims to develop and make broadly available a novel linear-modeling statistical methodology for analyzing sex-chromosome dosage compensation using genome-wide RNA-seq expression data. The statistical approaches currently employed for such analyses are far from ideal given the nature of the data and the desired set of inferences. Currently, biological replicates are averaged into a single measurement per gene and heavily normalized. Then particular effects of gene expression on the sex-chromosome relative to autosomes are evaluated using absolute expression while gene dosage effects are assessed using expression ratios, in both cases using non-parametric statistical tests. A more statistically robust approach is to employ linear mixed-effects modeling of gene expression. This provides a unified statistical framework to assess magnitude and significance of both chromosome-specific and dosage effects on gene expression. Moreover, it is applied directly to the sequencing read counts for each gene and incorporates scaling factors such as sequencing depth and transcript length into the models describing the data, as is done in most analyses of differential expression. As such, extensive normalization is avoided and statistical replicates are readily incorporated into the analysis. These methods will be implemented in a new software package, named "doseR", written in the R statistical programming language and distributed as part of the Bioconductor suite of bioinformatic software tools. Performance of the new statistical model and its software implementation relative to previous methods of assessing dosage compensation will be evaluated through extensive simulations of RNA-sequencing data. Application to specific empirical data sets relevant to dosage compensation will also be examined and evaluated. While the immediate motivation for software development is dosage compensation analysis, the proposed methodology can be employed in any analytical scenario requiring the detection of directional shifts in expression for multiple, specific subsets of genes. It therefore provides a tool with broad utility in systems biology research. Status and results of this project can be found at https://walterslab.github.io/doseR/.