NB Regression and HOA Inference for RNA-Seq Gene Expression Analysis

描述(申请人提供)：生物学家正在迅速采用RNA测序(RNA-Seq)来研究转录本，以基本了解细胞功能，并满足食品生产、食品安全、制药、人类健康、疾病治疗和疾病预防等领域的重要需求。然而，对RNA-Seq数据进行全面和具体分析的统计工具出现得很慢，而且使用为其他应用开发的现成工具很有可能产生误导性的结论。与这项建议的目标密切相关的是，用于评估来自RNA-Seq的差异基因表达的复杂方法-基于两组比较的负二项(NB)精确检验-尚未扩展到回归分析。这些方法是在考虑协变量后评估差异基因表达，分析表达对解释变量的依赖性，以及研究多个因素对表达的交互影响所必需的。该建议的目的是通过以下方式满足这一需要：1)开发、评估和实施对RNA-Seq数据的NB回归分析的似然比推断的高阶渐近(HOA)调整，包括准备用于RNA-Seq数据的完全回归分析的公开可用的R包，并将推断计算包括在已经公开可用的、基于Perl的计算管道中，用于完全分析RNA-Seq数据；2)阐明用于RNA-Seq研究的最佳推理的力量，并提供用于评估样本量需求的计算机程序；以及3)开发一个交互的、动态的可视化程序，用于传达RNA-Seq数据、NB回归模型结果和相关的不确定性。所使用的方法包括应用高阶渐近理论、蒙特卡罗模拟、发展细节级别(LOD)“焦点加上下文”可视化方法，以及认真关注真实的RNA-Seq数据集。

DESCRIPTION (provided by applicant): Biologists are rapidly adopting RNA-Sequencing (RNA-Seq) to study transcriptomes for basic understanding of cellular functions and to address important needs in such areas as food production, food security, pharmacy, human health, disease treatment, and disease prevention. Statistical tools for complete and specific analysis of RNA-Seq data, however, have been slow to emerge, and the use of off-the-shelf tools developed for other applications has the strong potential to produce misleading conclusions. Germane to the goals of this proposal, sophisticated methods for assessing differential gene expression from RNA-Seq- based on a negative binomial (NB) exact test for two-group comparisons-have not yet been extended to regression analysis. Such methods are required for assessing differential gene expression after accounting for covariates, for analyzing the dependence of expression on explanatory variables, and for studying interactive effects on expression of multiple factors. The objectives of this proposal are to address this need in the following ways: 1) develop, assess, and implement higher-order asymptotic (HOA) adjustments to likelihood ratio inference for NB regression analysis of RNA-Seq data, including the preparation of a publicly-available R package for complete regression analysis of RNA-Seq data, and the inclusion of the inferential computations in an already publicly available, Perl-based computational pipeline for complete analysis of RNA-Seq data; 2) clarify the power of optimal inference for RNA-Seq studies and provide a computer program for assessing sample size needs; and 3) develop an interactive, dynamic visualization program for conveying RNA-Seq data, NB regression model results, and associated uncertainties. The methods used include the application of higher-order asymptotic theory, Monte Carlo simulation, the development of Level of Detail (LOD) "focus plus context" visualization methods, and serious attention to real RNA-Seq datasets.