Transcriptome assembly and isoform expression level estimation from biased RNA-Seq reads
Transcriptome assembly and isoform expression level estimation from biased RNA-Seq reads
复制标题
DOI：
10.1093/bioinformatics/bts559
复制
发表时间：
2012-11-15
期刊：
影响因子：
5.8
通讯作者：
Jiang, Tao
中科院分区：
文献类型：
--
作者：
Li, Wei;Jiang, Tao
动机：RNA-Seq使用高通量测序技术，以前所未有的高分辨率和低成本鉴定和定量转录组。然而，RNA-Seq读数通常不是均匀分布的，并且RNA-Seq数据中的偏差在许多应用中提出了巨大的挑战，包括转录组组装和基因或同种型的表达水平估计。在文献中已经作出了很大的努力来校准有偏见的RNA-Seq数据的表达水平估计，但偏见对转录组组装的影响仍然在很大程度上unexplored.Results：在这里，我们提出了一个统计框架，从有偏见的RNA-Seq数据的转录组组装和亚型表达水平估计。使用准多项分布模型，我们的方法能够捕获各种类型的RNA-Seq偏差，包括位置，测序和可映射性偏差。我们在模拟和真实的RNA-Seq数据集上的实验结果显示了RNA-Seq偏差对转录组组装和亚型表达水平估计的有趣影响。我们的方法的优势是清楚地显示在实验分析中，其高灵敏度和精度的转录组组装和其估计的表达水平与定量逆转录-聚合酶链反应数据的高度一致性。
Motivation: RNA-Seq uses the high-throughput sequencing technology to identify and quantify transcriptome at an unprecedented high resolution and low cost. However, RNA-Seq reads are usually not uniformly distributed and biases in RNA-Seq data post great challenges in many applications including transcriptome assembly and the expression level estimation of genes or isoforms. Much effort has been made in the literature to calibrate the expression level estimation from biased RNA-Seq data, but the effect of biases on transcriptome assembly remains largely unexplored.Results: Here, we propose a statistical framework for both transcriptome assembly and isoform expression level estimation from biased RNA-Seq data. Using a quasi-multinomial distribution model, our method is able to capture various types of RNA-Seq biases, including positional, sequencing and mappability biases. Our experimental results on simulated and real RNA-Seq datasets exhibit interesting effects of RNA-Seq biases on both transcriptome assembly and isoform expression level estimation. The advantage of our method is clearly shown in the experimental analysis by its high sensitivity and precision in transcriptome assembly and the high concordance of its estimated expression levels with quantitative reverse transcription-polymerase chain reaction data.