6 min read
What is SignalP 6.0? Signal peptide prediction explained
SignalP 6.0 uses a protein language model to predict five signal-peptide classes and their cleavage sites without requiring the user to select an organism group.
Matic Broz Computational chemist
Key takeaways
- SignalP 6.0 predicts whether a protein has an amino-terminal signal peptide, which class it belongs to, and where it is likely to be cleaved.
- It supports five signal-peptide classes and does not require the user to choose an organism group.
- The fast model suits routine or high-throughput screening; the slow model adds detailed region annotations.
- A positive result predicts entry into a classical secretory pathway. It does not by itself prove that the mature protein is released outside the cell.
SignalP 6.0 is a machine-learning method for finding signal peptides at the amino terminus of protein sequences. These short sequence regions act like routing labels: they direct a newly synthesized protein into a secretion or membrane-translocation pathway and are often removed at a predicted cleavage site.
The method combines a pretrained protein language model with a conditional random field. This architecture lets SignalP 6.0 use patterns learned from large protein-sequence collections while producing residue-level predictions, including signal-peptide regions and cleavage sites.[1]
What does SignalP 6.0 predict?
For each submitted protein, SignalP reports whether an amino-terminal signal peptide is predicted, the likely signal-peptide class, and the most likely cleavage position. SignalP 6.0 supports all five signal-peptide types represented in its training and evaluation data:[1]
| Signal-peptide class | What it indicates |
|---|---|
| Sec/SPI | A standard Sec-pathway signal peptide cleaved by signal peptidase I |
| Sec/SPII | A Sec-pathway lipoprotein signal peptide cleaved by signal peptidase II |
| Sec/SPIII | A Sec-pathway pilin-like signal peptide cleaved by signal peptidase III |
| Tat/SPI | A twin-arginine translocation signal peptide cleaved by signal peptidase I |
| Tat/SPII | A twin-arginine lipoprotein signal peptide cleaved by signal peptidase II |
Earlier SignalP versions asked users to select an organism group. SignalP 6.0 instead infers the relevant sequence context from the protein itself, so the web service no longer requires that selection.[3]
How SignalP 6.0 works
Signal peptides are usually short and share broad biochemical features rather than one fixed sequence. A typical signal peptide contains:
- a positively charged amino-terminal region,
- a hydrophobic core,
- and a more polar region around the cleavage site.
SignalP's protein language model converts the sequence into contextual representations. The prediction layer then assigns residue-level labels and constrains them into biologically plausible signal-peptide regions. This is useful because it distinguishes a complete signal-peptide pattern from an isolated hydrophobic stretch elsewhere in the protein.[1]
The web service offers two model modes:
- Fast is the practical default for most analyses and for larger sequence sets.
- Slow provides more detailed signal-peptide region predictions when that extra annotation matters.
DTU recommends the fast model for most users and bulk prediction, while the slow model is intended for detailed analysis of signal-peptide regions.[3]
How to use SignalP 6.0 online
The SignalP 6.0 web service accepts protein sequences in FASTA format. A practical workflow is:
- Confirm that the input contains amino-acid sequences rather than nucleotide sequences.
- Give each FASTA record a short, unique identifier.
- Choose the fast model for routine screening or the slow model when detailed region labels are important.
- Submit the sequences and review both the protein-level class and the residue-level cleavage-site probabilities.
- Export the results and retain the SignalP version and model choice with the analysis.
For local or scripted analysis, DTU also distributes a portable Python package with the same core functionality as the web service. Academic users can request the package from DTU; the portable-package page provides the current access and licensing details.[4]
How to interpret a SignalP result
A positive prediction means the amino terminus resembles a signal peptide that can direct the protein into a classical secretory pathway. The predicted cleavage site marks the position where the signal peptide is most likely removed.
It does not necessarily mean the mature protein is secreted into the extracellular environment. In eukaryotes, proteins entering the endoplasmic-reticulum pathway may remain in the ER or Golgi, become membrane proteins, move to lysosomes, or be secreted. In bacteria, a signal peptide may direct a protein to the periplasm or a membrane rather than outside the cell. SignalP therefore predicts the entry signal, not the protein's final destination.[3]
Treat cleavage positions as hypotheses, especially when:
- several neighboring positions have similar probabilities,
- the amino terminus is unusually hydrophobic,
- the sequence may be incomplete,
- or a transmembrane helix could be mistaken for a signal peptide.
When the precise mature-protein boundary matters, compare the prediction with curated annotations, homologous proteins, mass-spectrometry evidence, or an experimental secretion assay.
What SignalP 6.0 does not predict
SignalP focuses on amino-terminal signals for the classical secretory route. It is not a general subcellular-localization predictor and does not cover every way a protein can leave a cell.[3]
Use another method or additional evidence when the question involves:
- Mitochondrial or chloroplast targeting peptides. These are transit peptides rather than the signal-peptide classes predicted by SignalP.
- Final subcellular location. A signal peptide does not distinguish extracellular proteins from proteins retained in secretory-pathway organelles or membranes.
- Leaderless secretion. Non-classical secretion can occur without an amino-terminal signal peptide.
- Internal signal anchors. A hydrophobic segment within a protein may control membrane insertion without being a cleaved amino-terminal signal peptide.
SignalP is best used as one part of an annotation workflow rather than as a standalone claim about protein localization or secretion.
Useful follow-up analyses
SignalP answers a targeted routing question. Other sequence properties can help check whether the result fits the rest of the protein:
- A hydropathy plot shows where hydrophobic segments occur across the full sequence and can reveal additional candidate membrane-spanning regions.
- The GRAVY calculator summarizes overall protein hydrophobicity.
- Protein Parameters reports molecular weight, theoretical isoelectric point, instability index, and other basic properties for the precursor or predicted mature sequence.
For cleavage-sensitive work, run these analyses both before and after removing the predicted signal peptide. That makes the effect of the proposed mature-protein boundary explicit.
Frequently asked questions
Does a positive SignalP result mean a protein is secreted?
No. It supports entry into a classical secretory pathway, but the protein may ultimately remain in an organelle, membrane, or bacterial periplasm. Final localization requires additional prediction or experimental evidence.[3]
Do I need to select an organism group in SignalP 6.0?
No. Unlike earlier versions, SignalP 6.0 does not require organism-group information from the user.[3]
Should I use the fast or slow SignalP model?
Use the fast model for most routine and high-throughput analyses. Use the slow model when detailed signal-peptide region annotations justify the additional runtime.[3]
Can SignalP find mitochondrial or chloroplast targeting peptides?
SignalP is designed for narrow-sense signal peptides, not general organelle-targeting or sorting signals. DTU recommends tools such as TargetP for mitochondrial and plastid transit peptides and broader localization tools when the final cellular destination is the question.[3]
Sources▼
- SignalP 6.0 predicts all five types of signal peptides using protein language models Nature Biotechnology · 2022. https://doi.org/10.1038/s41587-021-01156-3
- SignalP 6.0 web service DTU Health Tech · July 18, 2026. https://services.healthtech.dtu.dk/services/SignalP-6.0/
- SignalP 6.0 frequently asked questions DTU Health Tech · July 18, 2026. https://services.healthtech.dtu.dk/services/SignalP-6.0/5-FAQ.php
- SignalP 6.0 portable package DTU Health Tech · July 18, 2026. https://services.healthtech.dtu.dk/services/SignalP-6.0/7-Portable.php
Founder and computational chemist, ProteinIQ
Dr. Matic Broz is the founder of ProteinIQ and a computational chemist. He completed a PhD focused on protein structure, molecular dynamics, and neural networks, and writes about structural biology and scientific software.