miRNA-Seq normalization comparisons need improvement

Abstract

BACKGROUND Currently there is no method of best practice for the normalization of microRNA sequencing data (miRNA-Seq). Therefore, we read with interest a recent article in RNA by Garmire and Subramaniam that set out to compare various normalization strategies specifically for this application (Garmire and Subramaniam 2012). They compared methods currentlyinusefornormalizationofmessengerRNAsequencing (mRNA-Seq) data, such as total-depth normalization (“raw”) and Trimmed Mean of M-values (“TMM”). Additionally, they compared many methods not used previously with sequencing data, such as global scaling, and borrowed fromstrategies appliedtomicroarraystudies,such asquantile normalization (QN). The article attracted our attention for many reasons, but notably for the claimed poor performance and “abnormal results” of our TMM method (Robinson and Oshlack2010).Afterinvestigating,wediscoveredthatTMM’s claimedpoorperformancewastheresultofanerrorthatshiftedlog-ratiosinthewrongdirection.Furthermore,wefeltthat various practical issues were not satisfyingly discussed; we comment briefly on these here and provide reproducible reanalyses to support our claims (see Supplemental Material).