RNA reference materials and their corresponding reference datasets act as the 'ground truth' for the normalization of experimental values and are indispensable tools for reliably measuring intrinsically small differences in RNA-sequencing data, such as those between molecular subtypes of diseases in clinical samples. However, the variability in 'absolute' expression profiles measured across different batches, methods or platforms limits the use of conventional RNA reference datasets. We recently proposed a ratio-based method for constructing reference datasets. The ratio for a gene is defined as the normalized expression levels between two sample groups and produces more reliable values than the 'absolute' values obtained across diverse transcriptomic technologies and batches. Our gene ratios have been used for the successful generation of omics-wide reference datasets. Here, we describe a step-by-step process for establishing RNA reference materials and reference datasets, covering three stages: (1) reference materials, including material preparation, homogeneity testing and stability testing
(2) ratio-based reference datasets, including characterization, uncertainty estimation and orthogonal validation
and (3) applications, including definition of performance metrics, performing proficiency tests and diagnosing and correcting batch effects. This approach established the Quartet RNA reference materials and reference datasets (chinese-quartet.org) that have been approved as the first suite of nationally certified RNA reference materials by China's State Administration for Market Regulation. The protocol can be utilized to establish and apply reference materials to improve RNA-sequencing data quality in diverse clinical settings. The procedure can be completed in 2 d and requires expertise in molecular biology and bioinformatics.