sTarML: A Sequence-Based Machine Learning Framework for Predicting Bacterial sRNA-mRNA Interactions

sTarML prepares RefSeq genome + GTF for a given NCBI Taxonomy ID, builds start-codon–centered windows (covering 200 nt upstream and 100 nt downstream of the ATG), and predicts sRNA targets using bundled k-mer vectorizers/scalers and a trained model.

---

Install

# Install the package
conda install -c vkotsira starml

conda env create -f environment.yml
conda activate starml

---

Quick start

Comma-separated sequences

starml --taxid 511145 --srna "ACGTT,GGGAT" --outdir results --cores 4

FASTA of sRNAs

starml --taxid 511145 --srna-fasta my_sRNAs.fa --outdir results --cores 8

Specific RefSeq accession

starml --taxid 511145 --refseq GCF_025643435.1 --srna "ACGTT" --outdir results

Run starml --help for all options.

---

What it does (7 steps)

1. Initialize & logging → logs/run.log
2. Download genome + GTF (RefSeq) via ncbi-datasets-cli
3. Filter GTF to protein-coding mRNAs
4. Create start-codon–window BED (200 nt upstream, 100 nt downstream of ATG) and (if bedtools) extract FASTA
5. Normalize sRNAs (A/C/G/T/N; U→T) and save FASTA
6. Vectorize/scale, run bundled model → predictions.tsv
7. Finish

---

Outputs (in --outdir)

• startcodon_windows.bed (and startcodon_windows.fasta if bedtools present)
• srna/srnas.fasta
• predictions.tsv
• logs/run.log

---

Notes / troubleshooting

• --srna and --srna-fasta are mutually exclusive.
• If it stops after Step 5, you likely passed a filename to --srna; use --srna-fasta file.fa.
• “No assemblies found” → check --taxid or specify a valid --refseq GCF_....

---

Citation

A paper is forthcoming. Until then, please cite the software:

Vasiliki Kotsira et al.. sTarML: A Sequence-Based Machine Learning Framework for Predicting Bacterial sRNA-mRNA Interactions. Version 0.1.0.

@software{starml_software,
  author  = {Vasiliki Kotsira},
  title   = {sTarML: A Sequence-Based Machine Learning Framework for Predicting Bacterial sRNA-mRNA Interactions},
  year    = {2025},
  version = {0.1.0}
}

We’ll update this section (and CITATION.cff) with the final paper citation and DOI when available.

---

License

• Code: MIT (see LICENSE)
• Model files (starml/models/*): CC BY 4.0 (see LICENSE-models)