Fusion transcripts drive many cancers and contribute to heritable diseases, developmental disorders, and evolution. Third-generation RNA sequencing enables full-length transcript detection, but sequencing protocols and computational tools strongly influence accuracy.
We benchmarked four long-read fusion detection tools, JAFFAL, Genion, FusionSeeker, and LongGF, using simulated datasets and real cell-line data. We evaluated each tool for recall of known fusions, false positives, fusion type diversity, and breakpoint accuracy. We then applied the most precise and reliable tool to assess fusion detection across different long-read RNA sequencing methods.
To compare platforms and library preparations, we sequenced RNA from hepatocellular carcinoma cells (Huh7) using Oxford Nanopore (MinION) with PCR-cDNA, direct-cDNA, and direct-RNA protocols, and PacBio Revio with the Kinnex full-length RNA workflow. We analysed the datasets for fusion recovery, concordance, and technology-specific fusion profiles, focusing on fusions previously reported in Huh-7 cells.
Our results reveal how sequencing technology and analysis pipelines shape fusion detection outcomes. This study provides a clear evaluation of computational and experimental factors affecting fusion discovery and offers practical guidance for designing accurate long-read RNA sequencing studies.