A comprehensive assessment of RNA-seq accuracy, reproducibility and information content by the Sequencing Quality Control Consortium

performance of RNA-seq across laboratories and to test different sequencing platforms and data analysis pipelines. Here we report a multisite, cross-platform analysis of RNA-seq measurement performance in a controlled setting. We sequenced commercially available reference RNA samples spiked with synthetic RNA from the External RNA Control Consortium. Two distinct samples were assessed individually and also combined in known ratios. This allowed us to examine how well truths built into the study design, such as known relationships between samples within and across sites, could be recovered from measurements. With no independent ‘gold standard’ feasible, these ‘known truths’ support an objective assessment of performance. To this end, we examined a multitude of properties, including complementary metrics of reproducibility, accuracy and information content. Such a multidimensional characterization is critical for the development of more powerful analyses of the underlying biological mechanisms in complex data sets because often there is a trade-off between one desirable property and another, such as accuracy versus precision. Analyses focusing on measurement quality metrics11, spike-in controls and limits of detection47 and the effects of analytic pipeline choice are presented in separate studies (unpublished data). The SEQC project also involved studies assessing RNA-seq in several research applications (Fig. 1a), including a performance analysis of neuroblastoma outcome prediction (unpublished data), a comparative investigation of toxicogenomic samples testing chemicals with different modes of action12 and a comprehensive survey of tissue-specific gene expression in rat13. In total, >100 billion reads (10 terabases) of RNA-seq data were produced and studied, which to our knowledge represents the largest effort to date to generate and A comprehensive assessment of rNA-seq accuracy, reproducibility and information content by the sequencing Quality control consortium