The devil in the details of RNA-seq

Anton Kratz and Piero Carninci are in the Division of Genomic Technologies, Center for Life Science Technologies, RIKEN, Yokohama, Japan. e-mail: carninci@riken.jp or anton.kratz@riken.jp genes. Historically, northern blot analysis or quantitative real-time PCR have been the preferred techniques to measure gene expression with high precision and sensitivity. Their most glaring limitation, however, is that only a few genes can be measured at a time, and each single measurement is costly or too laborious to conduct on a transcriptome-wide scale. Microarrays were developed to measure simultaneously the expression of possibly all genes in information defining the phenotype of an organism, but knowledge of the transcriptome— the entirety of transcribed genes—is necessary to understand how the same genome can give rise to the many different cell types in an organism and how these genes are regulated in response to internal and external conditions, in health and in disease. Several techniques have been developed to detect and profile the expression of single RNA sequencing (RNA-seq) is a versatile and powerful tool for quantitatively measuring gene expression, detecting splice variants and singlenucleotide variations, and discovering novel coding and noncoding transcripts and fusion genes. Thus far, however, our understanding of the quantitative and qualitative characteristics of RNA-seq protocols and technologies has been lacking, especially in comparison to older, more established technologies such as microarrays. In this issue, the Sequencing Quality Control (SEQC)/Microarray Quality Control (MAQC)-III Consortium1–3 and the Association of Biomolecular Resource Facilities (ABRF)4 report results from systematic largescale experiments and analysis to elucidate the performance characteristics of RNA-seq and differences from previous technologies, different sources of measurement noise, and effects of data analysis and interpretation. These results will be of particular importance in the context of future projects on the scale of ENCODE5 or FANTOM6 as well as standardization of the data used by the larger community. In the future, it will become increasingly important to compare experimental data generated at different geographical sites using not only different library construction techniques but also supposedly identical measurement methods. The two main studies1,4 together with their companion publications in this issue enable informed decisions regarding study design, sample preparation, sequencing and data analysis protocols, and interpretation of future research. Whole-genome sequencing, first of simple unicellular organisms and now conducted almost routinely for higher organisms, has transformed the life sciences. The genome gives a static view of the genetic and regulatory The devil in the details of RNA-seq