Abstract
RNA-seq is a powerful tool for both discovery and experimentation. Most RNA-seq studies rely on library normalization to compare samples or to reliably estimate quantitative gene expression levels. Over the years a number of RNA-seq normalization methods have been proposed. Review studies testing these methods have provided evidence that commonly used methods perform well in simple normalization tasks, but their performance in challenging normalization tasks has yet to be evaluated. Here I test RNA-seq normalization methods using two challenging normalization scenarios. My assessment reveals surprising shortcomings of some commonly used methods and identifies an underappreciated method as the most promising normalization strategy for common, yet challenging RNA-seq experiments.