Explore double-stranded RNA structures across the human genome. dsRNAscan predicted 5,134,754 dsRNAs, mostly in repetitive and intergenic regions. Machine learning models trained on A-to-I editing and RNA structure-probing data identified ~2.4 million high-confidence predictions, which were enriched at dsRNA-binding protein binding sites.

Key Findings

• 5,134,754 predicted dsRNAs across the human genome, mostly in repetitive and intergenic regions
• ~2.4 million high-confidence by machine-learning models trained on A-to-I editing and RNA structure-probing data, enriched at dsRNA-binding protein binding sites
• 1,591 conserved across vertebrates
• 769 sense–antisense gene pairs forming intermolecular dsRNAs (GFF3 download)
• Ratio of intermolecular to intramolecular dsRNA correlates with ADAR dependency across cancer cell lines

The full 5,134,754 dsRNAs in two main formats: Parquet for analysis, BEDPE for genomic overlap.