Hierarchical Modeling and Parallelized Bayesian Inference for RNAseq Analysis

RNAseq 分析的分层建模和并行贝叶斯推理

• 批准号: 8639666
• 负责人: DANIEL NETTLETON
• 金额: $ 27.2万
• 依托单位: IOWA STATE UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2013
• 资助国家: 美国
• 起止时间: 2013-09-01 至 2017-05-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/8639666
• 关键词: Alleles; Bayesian Method; Bioinformatics; Biology; Complex; Computing Methodologies; Data; Data Analyses; Detection; Development; Dimensions; Disease; Exhibits; Funding; Gene Expression; Gene Expression Profiling; Genes; Genetic; Genetic Phenomena; Genomics; Hybrid Vigor; Hybrids; Inbreeding; Lead; Linear Models; Medical; Methods; Modeling; Molecular; Parents; Pattern; Public Health; RNA Sequence Analysis; RNA Sequences; Research; Research Personnel; Sample Size; Source; Technology; Work; complex biological systems; design; flexibility; gene function; offspring; public health relevance; response; statistics; tool

DESCRIPTION (provided by applicant): This proposal focuses on the development of hierarchical models and parallelized Bayesian inference for the analysis of RNA sequencing (RNAseq) data. Special emphasis is placed on gene expression profiling of parental inbred lines and their hybrid offspring for the discovery of key genes underlying heterosis, the genetic phenomenon otherwise known as hybrid vigor. The project will be led by a collaborative team of researchers with expertise in the analysis of high-dimensional gene expression data, Bayesian inference, bioinformatics, biology, computational methods, genetics, genomics, and statistics. The proposed research provides new tools for the analysis of high-dimension and low-sample-size count data generated by RNAseq technology. Hierarchical modeling allows for flexible information sharing across dimensions to extract as much information as possible from data. Parallel methods for Bayesian inference harness the power of modern computing to produce comprehensive results in a timely manner. Specific methods will be developed for (i) the identification of genes that exhibit expression heterosis, (ii) the detection of expressed and non-expressed genes, and (iii) the discovery of differential allele usage in hybrids. These methods will provide a deeper understanding of the molecular mechanisms of heterosis and lead to the discovery of key genes whose expression patterns provide hybrids with advantages over their parents. This information can be used to efficiently predict which of thousands of possible crosses will result in top performing hybrids. In addition to the specific methods mentioned above, hierarchical generalized linear models for the simultaneous analysis of tens of thousands of response variables will be developed. This work will permit the analysis of RNAseq data from complex designs with multiple sources of variability and will greatly extend the range of applicability for the funded research to encompass a variety of challenges in high-dimensional data analysis.

描述（由申请人提供）：该提案侧重于开发用于RNA测序（RNAseq）数据分析的分层模型和并行贝叶斯推理。 特别强调的是基因表达谱的亲本自交系和它们的杂交后代的关键基因的发现潜在的杂种优势，遗传现象，也被称为杂种优势。 该项目将由一个研究人员合作团队领导，该团队具有高维基因表达数据分析，贝叶斯推理，生物信息学，生物学，计算方法，遗传学，基因组学和统计学方面的专业知识。 该研究为分析RNAseq技术产生的高维和低样本量计数数据提供了新的工具。 分层建模允许跨维度灵活地共享信息，以便从数据中提取尽可能多的信息。 贝叶斯推理的并行方法利用现代计算的力量及时产生全面的结果。 具体的方法将被开发用于（i）鉴定表现出表达杂种优势的基因，（ii）检测表达和非表达基因，以及（iii）发现杂交种中的差异等位基因使用。 这些方法将提供一个更深入的了解杂种优势的分子机制，并导致发现的关键基因的表达模式提供杂种优势超过他们的父母。 这些信息可以用来有效地预测数千种可能的杂交中哪一种会产生最好的杂交种。除了上述具体方法外，还将开发用于同时分析数万个响应变量的分层广义线性模型。 这项工作将允许从具有多个可变性来源的复杂设计中分析RNAseq数据，并将大大扩展资助研究的适用范围，以涵盖高维数据分析中的各种挑战。