03-usage.md

SignalP 6.0 — Usage

Two ways to run it: the web server (no install, good for a handful of sequences) or the standalone Python package (for pipelines / large datasets).

Web server

URL: https://services.healthtech.dtu.dk/services/SignalP-6.0/
Paste sequences or upload a FASTA file.
Limits: proteins must be ≥10 aa; max 1000 proteins per job; long output "might timeout for more than 100 entries."
Organism: Eukarya (predicts Sec/SPI only) or Other (all five types).
Output format: Long (with per-residue probability figures) or Short (no figures).
Model mode: Fast or Slow (slow ≈ 6× slower, better region borders).
Overloaded? Use the BioLib mirror: https://dtu.biolib.com/SignalP-6

Standalone install

Download the package from the DTU "Downloads" tab (free for academics; commercial needs a license), then:

# in a fresh Python env (Python ≤ 3.10)
pip install signalp-6-package/
pip install "numpy<2"          # needed for the 6.0h build

# copy the bundled model weights into the installed package
SIGNALP_DIR=$(python3 -c "import signalp, os; print(os.path.dirname(signalp.__file__))")
cp -r signalp-6-package/models/* $SIGNALP_DIR/model_weights/

This exposes a signalp6 command inside that environment. (Source & instructions: https://github.com/fteufel/signalp-6.0)

Command line

signalp6 \
  --fastafile input.fasta \
  --organism other \
  --output_dir results/ \
  --format txt \
  --mode fast

| Argument | Short | Values | Default | Notes | |----------|-------|--------|---------|-------| | --fastafile | -ff | path | — | required input FASTA | | --output_dir | -od | path | — | required output folder | | --organism | -org | other, eukarya | other | eukarya → Sec/SPI only | | --format | -fmt | txt, png, eps, all, none | txt | plot/image outputs | | --mode | -m | fast, slow, slow-sequential | fast | speed vs. region accuracy / memory | | --bsize | -bs | int | 10 | batch size | | --torch_num_threads | -tt | int | 8 | CPU threads | | --write_procs | -wp | int | ≤ CPU count | parallel output writers | | --model_dir | -md | path | install default | custom weights |

Run signalp6 -h for the full list.

Output files

| File | Contents | |------|----------| | prediction_results.txt | Tab-delimited, one row per protein: ID, predicted type, per-type probabilities, cleavage-site position + probability | | processed_entries.fasta | Mature sequences (signal peptide removed) | | output.gff3 | SP start/end coordinates | | region_output.gff3 | n/h/c region coordinates | | output.json | Run parameters + paths to all outputs — the hook for pipeline integration | | *_plot.txt / *_plot.png / *_plot.eps | Per-position probabilities and figures (when --format requests them) |

Reading `prediction_results.txt`

The Prediction column is one of these labels (note it uses short codes, not the "Sec/SPI" names):

| Label | Meaning | |-------|---------| | OTHER | No signal peptide | | SP | Sec/SPI — standard secretory SP | | LIPO | Sec/SPII — lipoprotein SP | | TAT | Tat/SPI | | TATLIPO | Tat/SPII | | PILIN | Sec/SPIII — pilin |

Then there is one probability column per class (they sum to ~1) and a CS Position column giving the predicted cleavage site, e.g. CS pos: 21-22. Pr: 0.87 — cut between residues 21 and 22 with probability 0.87. The mature protein in processed_entries.fasta starts at that position.

Next: 04-benchmarks-and-context.md.