dsRNAscan

dsRNAscan is a bioinformatics tool for genome-wide identification of double-stranded RNA (dsRNA) structures. It uses a sliding window approach to detect inverted repeats that can form dsRNA secondary structures, with support for G-U wobble base pairing and ML-based scoring.

Browse human genome results at dsrna.chpc.utah.edu

Install

pip install dsrnascan

Linux and macOS, Python 3.8+ (Windows: use WSL).

Usage

# Scan a genome or transcriptome
dsrnascan input.fasta

# Scan a region
dsrnascan genome.fasta --only_seq chr21 --start 33455482 --end 33655482

By default dsRNAscan uses the settings of the human dsRNAome resource (--v1, on by default): windows of 10000 and 3000 nt, step 150, and a banded fold that gives ~99% identical results to the exact one. Any setting you change is listed in dsrnascan.log.

Run dsrnascan --help for all options.

Output

• *.dsrnas.parquet: one row per dsRNA with coordinates, search score, duplex fold, structure and ML scores (pd.read_parquet in Python, arrow::read_parquet in R; --txt writes the same table as tab-separated text)
• *.bedpe: paired arm coordinates (default)
• *.gff3: arms as linked features for IGV (--format gff3 or both)
• *.bp: base-pair arcs for IGV

ML scores (likely_edited, likely_forms) come from structure-only models trained on human data, where inverted Alus dominate. Treat them as a rough guide, especially outside Alus and in other species. Both scores run 0-1 (cutoffs: stability 0.553, probing 0.440).

Browse, filter and export

# Interactive browser: filters, sorting, structure view, export of the filtered set
dsrna-browse results_folder/

# With RNA editing sites (BED or GFF3), counted per arm and shown on the structure
dsrna-browse results_folder/ --editing-file editing_sites.bed

# The same filters from the command line
dsrna-browse results_folder/ --min percent_paired 85 --max "dG(kcal/mol)" -150 --export strong.gff3
dsrna-browse results_folder/ --region chr19:1-3000000 --likely-edited --export region.tsv

Export formats: .tsv, .parquet, .bedpe, .bed, .gff3, .bp, .fa, .dbn. --columns lists the filterable columns; --where takes an SQL condition on them.

The browser also has analysis tabs on the filtered set (distributions, scatter, dsRNA density along chromosomes, group summaries, statistics; charts download as PNG/SVG, their numbers as TSV) and an IGV tab showing the dsRNAs as arcs on a hosted assembly (--genome dm6, guessed from the results name) or the scan's own FASTA (--genome genome.fa).

Citation

If you use dsRNAscan, please cite:

dsRNAscan maps human dsRNAome, revealing conservation, intermolecular dsRNA, and correlates of ADAR dependency. Molecular Cell (2026). https://doi.org/10.1016/j.molcel.2026.08.001

License

GNU General Public License v3.0 - see LICENSE. Issues: GitHub Issues

Acknowledgments

dsRNAscan's compiled engine reimplements:

• EMBOSS einverted (inverted-repeat search): Rice P, Longden I, Bleasby A (2000) Trends Genet 16:276-277
• ViennaRNA duplexfold (duplex folding): Lorenz R et al. (2011) Algorithms Mol Biol 6:26, with Turner 2004 energy parameters (Mathews DH et al. (2004) PNAS 101:7287)