Scientific background

Data set

The data used for SignalP version 3.0 were extracted from SWISS-PROT version 40. The data sets were divided into prokaryotic and eukaryotic entries, and the prokaryotic data sets were further divided into Gram-positive eubacteria (Firmicutes) and Gram-negative eubacteria (Gracilicutes), excluding Mycoplasma and Archae. Viral, phage, and organellar proteins were not included. Negative training examples were taken from the SignalP version 2.

Extraction

In SWISS-PROT sequences with 'SIGNAL' in the FT line were extracted. Only sequence entries with experimental evidence were used and signal peptide entries marked 'HYPOTHETICAL', 'POTENTIAL', 'PROBABLE' were removed. Furthermore were entires with more than one cleavage site removed. From secretory proteins, the sequence of the signal peptide and the first 30 amino acids of the mature protein were included in the data set. From cytoplasmic and (for the eukaryotes) nuclear proteins, the first 70 amino acids of each sequence were used. Additionally, a set of eukaryotic signal anchor sequences, i.e. N-terminal parts of type II membrane proteins, were extracted.

Redundancy in the data sets was avoided by excluding pairs of sequences which were functionally homologous, i.e. where the cleavage site of one signal peptide could be located by simply aligning it to the other. The numbers of sequences remaining in the final data sets (version 3) are shown below:

Source Number of sequences
Signal peptides Cytoplasmic proteins Signal anchors Nuclear proteins
Eukaryotes 1192 459 67 990
Gram- 358 334 - -
Gram+ 153 151 - -

The redundancy reduction is described in detail in "Defining a similarity threshold for a functional protein sequence pattern: The signal peptide cleavage site", Henrik Nielsen, Jacob Engelbrecht, Gunnar von Heijne, and Søren Brunak, Proteins, 24, 165-177, 1996. The abstract can be found in the Abstract section

Data set cleanup

Isoelectric point

We were able to identify wrongly annotated signal peptides from Swiss-Prot simply by calculating the isoeletric point of the signal peptide and corresponding mature protein. Using this approach we were even able to identify wrongly annotated start codons. Indepth information on this issue can be found in "Improved prediction of signal peptides - SignalP 3.0 ", Bendtsen et al., 2004. The abstract can be found in the Abstract section.

Propeptide

In Swiss-Prot, annotated signal peptide cleavage sites are entries found do in some cases contain propeptides. These sequence entries were reannotated acording to the output of SignalP version 2. Propeptides have basic residues (R/K) at their cleavage sites.

Spurious residues at position -1 was also removed in the training set. Only amino acids A,L,G,xxx were alowed at this position.

Wrong start codons

In specific cases, SignalP can be used as a "validation" tool for correct startcodon annotation. In Swiss-Prot entry SFMA_ECOLI (Swiss-Prot rel. 40) the annotated Signal peptide cleavage site was positioned 22 and a predicted cleavage site at position at 34. As the S-score (indicating signal peptide-ness) is very low until position 11 we believed that the methionine at position 12 is the true start of the protein. In a later release of Swiss-Prot this was indeed verified.
Wrong start codon

Sequence logos

To visualize the sequence information content in signal peptides, we have generated sequence logos for the eukaryotic, Gram negative and Gram positive training set. The total height of the stack of letters at each position shows the amount of sequence conservation at the position, while the relative height of each letter shows the relative abundance of the corresponding amino acid.

Sequence logos

Sequence logos of signal peptides, aligned after their cleavage sites.

Blue: Positively charged residues
Red: Negatively charged residues
Green: Neutral polar residues
Black: Hydrophobic residues

Length distributions of signal peptides

Distribution of lengths of the eukaryotic and prokaryotic signal peptides. The average length is 22.6 amino acids for eukaryotes, 25.1 for Gram-negative bacteria, and 32.0 for Gram-positive bacteria.

Length distributions of signal peptides

Characteristics of signal peptides

The common structure of signal peptides from various proteins is commonly described as a positively charged n-region, followed by a hydrophobic h-region and a neutral but polar c-region. The (-3,-1)-rule states that the residues at positions -3 and -1 (relative to the cleavage site) must be small and neutral for cleavage to occur correctly.

We have analyzed the characteristics of the new signal peptide data set and displayed them in the form of sequence logos as previously shown. The differences between signal peptides from different organisms are apparent from the sequence logos:

Eukaryotes Prokaryotes
Gram-negative Gram-positive
Total length (average) 22.6 aa 25.1 aa 32.0 aa
n-regions only slightly Arg-rich Lys+Arg-rich
h-regions short, very hydrophobic slightly longer, less hydrophobic very long, less hydrophobic
c-regions short, no pattern short, Ser+Ala-rich longer, Pro+Thr-rich
-3,-1 positions small and neutral residues almost exclusively Ala
+1 to +5 region no pattern rich in Ala, Asp/Glu, and Ser/Thr

Download of dataset

The datasets used for the current version of SignalP are not yet publicly available. The old datasets can still be downloaded through here.