Transcriptome Analysis and Applications Based on Next-generation RNA Sequencing Data.

摘要

The recent development of next generation RNA-sequencing, termed 'RNA-Seq', has offered an opportunity to explore the RNA transcripts from the whole transcriptome. As a revolutionary method, RNA-Seq not only could precisely measure the abundances of transcripts, but discover the novel transcribed contents and uncover the unknown regulatory mechanisms. Meanwhile, the combination of different levels of next-generation sequencing, such as genome sequencing and methylome sequencing has provided a powerful tool for novel discovery in the biological context.;My PhD study focuses on the analysis of next-generation sequencing data, especially on RNA-Seq data. It mainly includes three parts: pipeline development analysis, data analysis and mechanistic study.;As the next-generation sequencing (NGS) technology, the analysis of massive NGS data is a great challenge. Many existing general aligners (as contrast to splicing-aware alignment tools) are capable of mapping millions of sequencing reads onto a reference genome. However, they are neither designed for reads that span across splice junctions (spliced reads) nor for reads that could match multiple locations along the reference genome (multireads). Hence, we have developed an ab initio mapping method - ABMapper, using two-seed strategy. The benchmark results show that ABMapper can get higher accuracy and recall compared with the same kind of tools: TopHat and SpliceMap. On the other hand, the selection of the most probable location for spliced reads and multireads becomes a big problem. These reads are randomly assigned to one of the possible locations or discarded completely when calculating the expression level, which would bias the downstream analysis, such as the differentiated expression analysis and alternative splicing analysis. To rationally determine the location of spliced reads and multireads, we have proposed a maximum likelihood estimation method based on a geometric-tail (GT) distribution of intron length. This probabilistic model deals with splice junctions between reads, or those encompassed in one or both of a pair-ended (PE) reads. Based on this model, multiple alignments of reads within a PE pair can be properly resolved.;The accumulation of NGS data has provided rich resources for deep discovery of biological significance. We have integrated RNA-Seq data and methylation sequencing data to build a predictive model for the regulation of gene expression based on DNA methylation patterns. We found that DNA methylation could predict gene expression fairly accurately and the accuracy can reach up to 78%. We have also found DNA methylation at gene body is the most important region in these models, even more useful than promoter. Finally, feature overlap network based on an optimum subset of combination of all methylation patterns and CpG patterns has indicated the collaborative regulation of gene expression by DNA methylation patterns.;Not only new algorithms were developed to facilitate the RNA-Seq data analysis, but the transcriptome analysis was performed on zebrafish. The analysis of differentially-expressed genes and pathways involved after calycosin treatment, combined with other experimental evidence such as fluorescence microscopy and quantitative real-time polymerase chain reaction (qPCR), has well demonstrated the proangiogenic effects of calycosin in vivo.;In summary, this thesis detailed my work on NGS data analysis, discovery of biological significance using data-mining algorithms and transcriptome analysis.