Statistical methods for elucidating regulatory mechanisms and functional impacts of transcriptome variation at population and single-cell scales

点击翻译按钮获取中文摘要

PROJECT SUMMARY / ABSTRACT Bulk RNA sequencing (RNA-seq) and single-cell RNA sequencing (scRNA-seq) are powerful high-throughput techniques for studying transcriptome variation at population and single-cell scales. Many computational methods have been developed for analyzing bulk RNA-seq and scRNA-seq data. However, there remain multiple challenges in identifying disease/trait-associated genes from population-scale bulk RNA-seq data, studying temporal transcriptome dynamics from scRNA-seq data, and benchmarking scRNA-seq computational tools. In our proposed research, we will develop statistical methods to address these challenges and elucidate regulatory mechanisms of transcriptome variation at population and single-cell scales. At the population scale, we will develop a unified statistical framework for identifying associations between genotypes and RNA isoform abundances, the “ideal” RNA-level molecular phenotypes. Our framework will unify existing diverse approaches that focus on specific aspects of transcript variation (e.g., gene expression, alternative exon/intron usage, and alternative polyadenylation) and, for the first time, incorporate the uncertainty in estimating isoform abundances. As a result, our framework should improve the accuracy and power in detecting associations between genetic variants and genes. We will make our framework applicable to all second- and third-generation RNA-seq data and apply it to the GTEx data, the most comprehensive genotype-transcriptome database, to discover genes that are associated with the disease/trait-associated variants found by GWAS. At the single-cell scale, we will develop three methods: 1) a valid statistical test for detecting temporally differentially expressed genes from scRNA-seq data while accounting for the uncertainty in trajectory inference, 2) a clustering method that integrates mechanistic and statistical modeling for identifying cell subpopulations along a temporal process, and 3) a comprehensive and interpretable simulator that generates realistic scRNA-seq data for benchmarking computational tools. The first two methods will offer much-in-demand solutions to temporal gene expression analysis of scRNA-seq data. Their applications will include the study of macrophage transcriptome changes during immune responses. The third method will be the first scalable and transparent simulator that captures gene correlations and allows the tuning of experimental parameters, including cell numbers and library sizes. Overall, we expect that our proposed methods will significantly improve the power, robustness, and reproducibility of studying transcriptome variation from bulk and single-cell RNA-seq data.