Statistical methods for improving reproducibility and utility of sequencing data

描述（由申请人提供）：基于测序的测定已成为研究全基因组蛋白质-DNA相互作用和由组蛋白修饰（均通过ChIP-seq）以及转录组（RNA-seq）定义的染色质状态的首选技术。尽管其广泛使用，许多实验和数据分析的挑战仍然必须克服，以达到可靠的和可重复的数据的生物学解释。每项研究的样本量都很小，进一步限制了数据分析的效力和可靠性。当来自不同研究的重复样品或相似样品可用时，重复样品之间的再现性告知我们鉴定的保真度，并且可能用于检测在单个样品中太温和而不能可靠检测的可再现信号。我们建议开发一套新的统计方法，利用重复样品提供的再现性信息来检查实验质量，选择可靠的鉴定，并优化实验设计中的操作参数。目标1将开发统计方法来评估鉴定的重复性，并通过几种基于测序的分析中的重复性来选择鉴定。基于可重复性的选择标准补充了单个样本的通常显著性度量，但具有跨数据集，平台和不同显著性度量的可比性。目标2将开发一个回归框架，以评估实验和数据分析程序中的操作参数如何影响ChIP-seq和RNA-seq实验的重现性。它将允许人们表征协变量对测定重现性的同时和独立影响，并在控制潜在混杂变量的同时比较方案的重现性。目标3将开发半参数，基于秩的荟萃分析方法，用于整合来自不同来源的基于RNA-seq的转录组分析。所提出的方法将考虑到由于数据来源的异质性，并且将在荟萃分析中纳入研究目标。

DESCRIPTION (provided by applicant): Sequencing-based assays have become the technology of choice for studying genome-wide protein-DNA interactions and chromatin states defined by histone modifications (both by ChIP-seq) as well as transcriptomes (RNA-seq). Despite their widespread use, many experimental and data-analytical challenges still must be overcome to reach reliable and reproducible biological interpretations of the data. The small sample size of each individual study further limits the power and reliability of data analyses. When replicate samples or similar samples from different studies are available, reproducibility across replicate samples informs us about the fidelity of the identification, and potentially it ca be used to detect reproducible signals that are too modest to be detected reliably in individual samples. We propose to develop a suite of new statistical methods that make use of the reproducibility information provided by the replicate samples to examine the quality of experiments, select reliable identifications, and optimize operational parameters in the experimental design. Aim 1 will develop statistical methods to assess the reproducibility of identifications and to select identifications by their reproducibility in several sequencing-based analyses. The reproducibility-based selection criterion complements the usual measure of significance on a single sample, but has the benefit of being comparable across data sets, platforms and different measures of significance. Aim 2 will develop a regression framework to assess how operational parameters in the experimental and data analytical procedures affect the reproducibility of ChIP-seq and RNA-seq experiments. It will allow one to characterize the simultaneous and independent effects of covariates on reproducibility of the assays and to compare reproducibility of protocols while controlling for potential confounding variables. Aim 3 will develop semi-parametric, rank-based meta-analysis methods for integrating RNA-seq-based transcriptome analyses from different sources. The proposed methods will take into account heterogeneity due to data sources, and they will incorporate the study goals in the meta-analysis.