|
Abstract:
|
It was initially claimed that RNA-seq produced unbiased data that did not require sophisticated normalization. However, studies have since revealed that RNA-seq data is biased and that accurate detection of differentially expressed genes is not a trivial task. In light of these findings, researchers should analyze RNA-seq data like they would any other biased multivariate data. The most effective approach to modern data analysis is to iterate between models and visuals, and to enhance the appropriateness of models based on feedback from visuals. Unfortunately, researchers do not often use models and visuals in a complimentary fashion when analyzing RNA-seq data. Here, we use real RNA-seq data to show that our visualization tools can detect normalization problems, DEG designation problems, and common errors in RNA-seq analysis. We also show that our tools can identify genes of interest that cannot be obtained by models. In this paper, we do not propose that users radically change their approach to RNA-seq analysis. Instead, we propose that users simply modify their approach to RNA-seq analysis by assessing the sensibility of their models with multivariate statistical graphics.
|
