Benchmarking RNA-seq with the Quartet Reference Materials to establish Best Practices for Accurate Alternative Splicing Detection

Previous limited characterization of RNA-seq accuracy in alternative splicing (AS) due to methodological diversity and lack of reference standards has left unclear how to achieve optimal performance for informed applications—an issue increasingly critical with the rise of long-read sequencing. To address this, we conducted the first large-scale reference-based benchmarking of AS detection accuracy across 42 laboratories and 159 analysis pipelines, leveraging the Quartet reference materials. We show that high-quality RNA-seq enables relatively accurate isoform detection. Best practices for experimental and bioinformatic designs were identified, and achieved mean Pearson correlations of 0.82 for isoform quantification and Matthews correlation coefficients of 0.69 for differential expression analysis, representing improvements of 0.21–0.31 and 0.46–0.61 across laboratories, respectively, compared to the poorest workflows. AS event-level accuracy remained limited, yet the best tools still outperformed the poorest by 0.02–0.25 in event quantification and 0.10–0.25 in differential splicing analysis. Beyond technical variables, low expression levels were the primary constraint on isoform and event detection accuracy, followed by their compositional complexity. Collectively, this study provides practical guidance for maximizing AS profiling accuracy with existing methodologies, contributing to effective RNA-seq application in splicing research.