Edit · 01 Overview

Saving needs a token. Log in or register to get one. Edits are public and licensed CC-BY-SA-4.0.

# SignalP 6.0 — Overview

## What is a signal peptide?

A **signal peptide (SP)** is a short stretch of amino acids (typically ~15–30
residues) at the **N-terminus** of a protein that acts as a molecular "address
label." It directs the nascent protein to the secretory pathway / translocation
machinery in the membrane, and is then **cleaved off** by a signal peptidase,
releasing the mature protein. Getting SP prediction right matters for:

- Annotating secretomes and membrane/surface proteins
- Designing recombinant proteins for secretion (biotech, antibody/enzyme production)
- Finding vaccine and drug targets (secreted/surface antigens)
- Metagenomics, where the source organism is unknown

Classic SP structure has three regions, all of which SignalP 6.0 predicts:

| Region | Property | Role |
|--------|----------|------|
| **n-region** | Positively charged (basic residues) | Interacts with the translocon / membrane surface |
| **h-region** | Hydrophobic core | Inserts into the membrane |
| **c-region** | Polar, ends in the cleavage site (often A-x-A motif) | Recognized by signal peptidase |

> A key caveat printed on the DTU page: *"the presence or absence of a signal
> peptide is not the whole story about localization."* SignalP tells you whether
> a protein enters the secretory pathway, not its final destination — use
> **DeepLoc / DeepLocPro** for subcellular localization and **NetGPI** for GPI anchors.

## The five signal-peptide types

The headline feature of v6.0 is that it discriminates **all five** SP types found
across Archaea and Bacteria (Eukarya only have Sec/SPI). Each type is defined by
its **translocon** (how the protein crosses the membrane) and its **signal
peptidase** (which enzyme cuts it).

| Type | Output label | Translocon | Peptidase | Distinguishing feature |
|------|--------------|-----------|-----------|------------------------|
| **Sec/SPI**  | `SP`      | Sec | SPase I (Lep/LepB)   | "Standard" secretory SP; full n/h/c structure |
| **Sec/SPII** | `LIPO`    | Sec | SPase II (Lsp/LspA)  | **Lipoprotein** SP; conserved **lipobox** with an invariant cysteine at the +1 position |
| **Tat/SPI**  | `TAT`     | Tat | SPase I (Lep)        | **Twin-arginine (S/T-R-R-x-F-L-K)** motif in the n-region; transports folded proteins |
| **Tat/SPII** | `TATLIPO` | Tat | SPase II (Lsp)       | Twin-arginine motif **and** a lipobox (Tat lipoprotein) |
| **Sec/SPIII**| `PILIN`   | Sec | SPase III (PilD/PibD)| **Pilin / pilin-like** SPs (type IV pili); cleaved within the h-region, no classic c-region |

- **Sec vs Tat:** the **Sec** translocon exports unfolded chains; the **Tat**
  (twin-arginine translocation) pathway exports already-folded proteins and is
  marked by the conserved twin-arginine motif.
- **SPase I / II / III:** three different signal peptidases produce the three
  cleavage chemistries. SPase II (lipoproteins) leaves a lipid-modified cysteine;
  SPase III processes pilins.

SignalP **5.0** could only reliably detect **three** types (Sec/SPI, Sec/SPII,
Tat/SPI). Version 6.0 adds **Tat/SPII** and **Sec/SPIII**, which had too few known
examples to model before the protein-language-model approach.

## Why 6.0 is a step change

1. **All five types in one pass**, including the two rare ones.
2. **Organism-agnostic**: works on sequences of **unknown origin** (metagenomic /
   metatranscriptomic data), because the language model generalizes rather than
   relying on organism-specific statistics.
3. **Region decoding**: returns not just yes/no + cleavage site but the internal
   **n/h/c architecture** of the SP.
4. **Better generalization to distant homologs**: at <60% identity it clearly
   outperforms SignalP 5.0.

See [02-architecture.md](02-architecture.md) for how the model achieves this and
[04-benchmarks-and-context.md](04-benchmarks-and-context.md) for the numbers.

Cancel

Preview · live