Long-read RNA sequencing data for the identification of fusion transcripts in rice

Fusion transcripts, arising from genomic rearrangements or RNA-level events like trans-splicing and readthrough transcription, are increasingly observed in plant transcriptomes. However, the authenticity of many fusion transcripts remains a significant concern, compromised by confounding factors such as technical artifacts and annotation errors, and systematic investigations in plants are limited. Here, we comprehensively evaluate fusion transcripts in Oryza sativa using multiple long-read RNA sequencing platforms. Through cross-platform comparisons, breakpoint sequence analysis and experimental validation, we demonstrate that the vast majority of detected fusions, particularly those without genomic support, are false positives. These spurious events originate from various hidden sources, including short homologous sequence-mediated template switching, alignment errors, mis-annotation, and reference genome bias. Based on those findings, we propose a robust and stepwise framework for classifying and filtering false-positive fusion transcripts, integrating sequencing platform selection, breakpoint features, reference-aware mapping, and control-based validation. This framework enables the reliable elimination of false-positive fusion events, providing a critical foundation for further accurate functional investigation of bona fide fusion transcripts in plant.