SignalP - 4.1

Submission

---

Sequence submission: paste the sequence(s) and/or upload a local file

Paste a single sequence or several sequences in FASTA format into the field below:

Submit a file in FASTA format directly from your local disk:

Organism group Eukaryotes Gram-negative bacteria Gram-positive bacteria	D-cutoff values Default (optimized for correlation) Sensitive (reproduce SignalP 3.0's sensitivity) User defined: D-cutoff for SignalP-noTM networks D-cutoff for SignalP-TM networks	Graphics output No graphics PNG (inline) PNG (inline) and EPS (as links)
Output format Standard Short (no graphics) Long All - SignalP-noTM and SignalP-TM output (no graphics)	Method Input sequences may include TM regions Input sequences do not include TM regions	Positional limits Minimal predicted signal peptide length. Default: 10 N-terminal truncation of input sequence (0 means no truncation). Default: Truncate sequence to a length of 70 aa

---

Restrictions:
At most 2,000 sequences and 200,000 amino acids per submission; each sequence not more than 6,000 amino acids.

Confidentiality:
The sequences are kept confidential and will be deleted after processing.

---

CITATIONS

For publication of results, please cite:

SignalP 4.0: discriminating signal peptides from transmembrane regions
Thomas Nordahl Petersen, Søren Brunak, Gunnar von Heijne & Henrik Nielsen
Nature Methods, 8:785-786, 2011

doi: 10.1038/nmeth.1701
PMID: 21959131
Supplementary materials: nmeth.1701-S1.pdf

Other relevant papers:

• Original paper (version 1.0):

  Identification of prokaryotic and eukaryotic signal peptides and prediction of their cleavage sites.
  Henrik Nielsen, Jacob Engelbrecht, Søren Brunak and Gunnar von Heijne.
  Protein Engineering, 10:1-6, 1997.

• SignalP-HMM (version 2.0):

  Prediction of signal peptides and signal anchors by a hidden Markov model.
  Henrik Nielsen and Anders Krogh.
  Proceedings of the Sixth International Conference on Intelligent Systems for Molecular Biology (ISMB 6),
  AAAI Press, Menlo Park, California, pp. 122-130, 1998.

• Version 3.0:

  Improved prediction of signal peptides: SignalP 3.0.
  Jannick Dyrløv Bendtsen, Henrik Nielsen, Gunnar von Heijne and Søren Brunak.
  J. Mol. Biol., 340:783-795, 2004.

  Download the full article in PDF.

• Paper about using SignalP and other protein subcellular localization prediction methods:

  Locating proteins in the cell using TargetP, SignalP, and related tools
  Olof Emanuelsson, Søren Brunak, Gunnar von Heijne, Henrik Nielsen
  Nature Protocols 2:953-971 (2007).
  

  Access the paper and supplementary materials.

Frequently Asked Questions

Changes from version 4.0 to 4.1
Changes from version 3 to 4
Biological background, signal peptides
Biological background, other sorting signals
Biological background, organism groups
History

---

Changes from version 4.0 to 4.1

— What's new?

Please see the Version history.

— Why do you present a choice between two cutoff settings? Can't you just decide on one?

The optimal cutoff really depends on what you want to use the method for. If it is important to find all signal peptides, use the sensitive cutoff. If you want an estimate of the number of signal peptides in a genome, use the default cutoff.

— Why have you imposed a minimum length?

Because we believe that predictions of signal peptides shorter than ten residues made by SignalP 4.1 are false. The shortest known signal peptides are 11 residues long (with one exception, SP23_TENMO, which does not look like a signal peptide at all). Click here for an updated list of experimentally confirmed signal peptides from UniProt of length 11 or shorter.

— What happened to the Background page?

It's here! The important material from the Background page has been integrated into this FAQ, we hope you like the new format.

---

Changes from version 3 to 4

— What's new?

Please see the Version history.

— What happened to the HMM part?

While making SignalP 4.0, we did retrain the Hidden Markov Model (HMM) part of SignalP. However, we found that it did not perform better than the neural networks in any of the performance parameters we tested. Therefore, we decided not to include it. If the HMM output is important for you, you can still use SignalP 3.0.

— Why is my favourite signal peptide no longer predicted correctly? SignalP 3.0 could do it!

As explained on the Performance page, SignalP 4 with the default cutoff has a lower sensitivity than SignalP 3. Please try again with the new "Sensitive" setting.

— What happened to the Yes/No answers for max C score etc.?

SignalP 3.0 provided five Yes/No answers for the NN part. We found that this was confusing for users and obscured the fact that the D-score is the best score for discriminating between signal peptides and non-signal peptides.

---

Biological background, signal peptides

— What are signal peptides?

The term "signal peptide" is used with two meanings: In the broad sense (used in many textbooks), a signal peptide is any sorting signal embedded in the amino acid sequence of a protein. In the narrow sense (used in most of the scientific literature), a signal peptide is an N-terminal signal that directs the protein across the ER membrane in eukaryotes and across the plasma membrane in prokaryotes. Signal peptides in the narrow sense are also known as ER signal peptides or secretory signal peptides. Read more in UniProt, in Wikipedia, and in the Sequence feature ontology.

It is important to emphasize that SignalP predicts signal peptides in the narrow sense only.

— Are signal peptides always N-terminal?

In the narrow sense: Yes, per definition. In the broad sense: No, there are several sorting signal that are C-terminal (e.g. the PTS1 signal for peroxisomal import) or internal (e.g. the nuclear localization signal).

— Are signal peptides (in the narrow sense) always cleaved?

No, there are rare cases of uncleaved signal peptides. For an updated list of such proteins annotated in UniProt, click here. These should not be confused with signal anchors, see below. In SignalP, these typically get high S-scores but low C- and Y-scores.

— Which protease is responsible for signal peptide cleavage?

In bacteria, it is Signal Peptidase I (SPase I), also known as Leader Peptidase (Lep). In eukaryotes, it is the signal peptidase complex (SPC), which consists of four subunits in yeast and five in mammals. Read more in MEROPS.

— My protein has a signal peptide. Can I then safely conclude that it is secreted?

No. You can only conclude that it enters the secretory pathway.

In eukaryotes, there are several opportunities for a protein with a signal peptide to escape secretion. It could:

• It could be retained in the endoplasmic reticulum (ER). Soluble ER-resident proteins have a C-terminal retention signal with the consensus sequence KDEL, see PROSITE.
• be retained in the Golgi apparatus,
• be directed to the lysosome (vacuole in plants and fungi),
• have one or more transmembrane helices and therefore be retained in either the plasma membrane, or one of the membranes of the secretory pathway (ER, Golgi, lysosome/vacuole), or
• have a signal for GPI-anchoring, a C-terminal cleaved peptide which functions as a signal for attachment of a Glycophosphatidylinositol (GPI) group that anchors the protein to the outer face of the plasma membrane.

In Gram-positive bacteria, a protein with a signal peptide could:

• have one or more transmembrane helices, or
• be attached to the cell wall.

In Gram-negative bacteria, a protein with a signal peptide could:

• have one or more transmembrane helices,
• be retained in the periplasm, or
• be inserted into the outer membrane as a β-barrel transmembrane protein.

— Does SignalP predict signal peptides of bacterial lipoproteins?

No. Bacterial lipoproteins have special signal peptides which are cleaved by Signal Peptidase II (SPase II), also known as Lipoprotein signal peptidase (Lsp). A diacylglyceryl group is attached to a Cysteine residue in position +1 relative to the cleavage site, which bears no resemblance to the SPase I cleavage site. See also MEROPS and PROSITE.

For prediction of prokaryotic lipoproteins we recommend using the LipoP server.

— Does SignalP predict TAT (Twin-arginine translocation) signal peptides?

Not very well. Bacterial TAT signal peptides, which direct their proteins through an alternative translocon (TatABC), have a special motif containing two Arginines in the n-region. Additionally, they are in general longer and less hydrophobic than normal (Sec) signal peptides. As a consequence, they are sometimes missed by SignalP, even though they are cleaved by SPase I (Lep). See also PROSITE and InterPro.

For prediction of TAT signal peptides we recommend using the TatP server.

---

Biological background, other sorting signals

— What are signal anchors?

A signal anchor is a transmembrane helix located close to the N-terminus of a protein with an N-in orientation (i.e. the N-terminus is on the cytoplasmic side of the membrane). It functions much like a signal peptide since it is recognized by the Signal Recognition Particle (SRP) and inserted into the translocon; but instead of being cleaved and degraded it remains in the membrane and anchors the protein to it. Proteins anchored in this way are known as Type II transmembrane proteins.

Signal peptides (above) versus signal anchors (below)

It is important to realize that the difference between signal peptides and signal anchors is not a question of presence or absence of a cleavage site. Instead, the most important difference seems to be the length of the hydrophobic domain. It has been shown experimentally that it is possible to convert a cleaved signal peptide to a signal anchor merely by lengthening the h-region, without altering the cleavage site (Chou & Kendall 1990; Nilsson, Whitley, & von Heijne 1994).

The introduction of the Hidden Markov Model (HMM) method in SignalP version 2 made it possible to some extent to distinguish signal peptides from signal anchors (in that version, only in eukaryotes). However, SignalP 4 (based entirely on the Neural Network (NN) method), does a better job, since its negative set is not confined only to transmembrane helices annotated as signal anchors, but includes all types of transmembrane segments close to the N-terminus.

— What should I use for predicting signal peptides in the broad sense?

For mitochondrial and plastid import signals, also known as transit peptides, we recommend TargetP. For other sorting signals, we refer to our paper "Locating proteins in the cell using TargetP, SignalP, and related tools".

— What should I use for predicting non-classical (leaderless) secreted proteins?

Not all secretory proteins carry signal peptides. Some proteins enter a non-classical secretory pathway without any currently known sequence motif. In eukaryotes, these proteins are mostly growth factors and extracellular matrix binding proteins. In Gram-negative bacteria, the type I, III, IV and VI secretion systems function without signal peptides. For prediction of such proteins we recommend the SecretomeP server.

---

Biological background, organism groups

— Why is there no version for Archaea?

Because there are too few experimentally confirmed signal peptides from this organism group in the UniProt database (click here for an updated list).

— Which version should I use for vira and bacteriophages?

You should use the version corresponding to the host organism. There are some indications that viral signal peptides differ from those of the host organism, but SignalP currently does not take that into account.

— Which version should I use for Tenericutes/Mollicutes (Mycoplasma and related genera)?

You shouldn't use SignalP at all for these organisms, since they seem to lack a type I signal peptidase completely!

— Which version should I use for metagenomic sequences of unknown origin?

This is an unsolved question. Please use all three versions to search for signal peptides in such data.

— Is one version enough for all eukaryotic organisms, or are there differences within the eukaryotes?

It is known that some yeast signal peptides are not recognized by mammalian cells (Bird et al., 1987 and 1990). Therefore, it would be natural to assume that separate SignalP versions for yeast and Mammalia would provide better predictions than a common eukaryotic version. While developing SignalP 4.0 we tried dividing the eukaryotic data into animals, fungi, and plants and training separate methods for these three groups. However, this did not give any improvement, and performance for all three groups was better when using the method trained on all eukaryotic sequences together.

— Are two versions enough for all bacteria, or are there differences within the Gram-positive/Gram-negative bacterial groups?

The Gram-negative version of SignalP is almost certainly biased towards E. coli and other γ-proteobacteria, since these constitute the bulk of the experimentally annotated bacterial proteins in UniProt. Unpublished results suggest that some bacteria have very divergent cleavage site motifs. Future versions of SignalP might therefore divide the Gram-negative bacteria into several classes, if data are available.

Gram-positive bacteria probably constitute a more homogenous group, but it is an open question whether there are differences in signal peptides between Actinobacteria (high G+C Gram-positive bacteria) and Firmicutes (low G+C Gram-positive bacteria). More data on Actinobacteria are needed before that can be answered.

---

History

— How are the various versions of SignalP related?

Please see the Version history.

— Was there ever a Nobel prize awarded for signal peptides?

Yes, for signal peptides in the broad sense. The importance of signal peptides was emphasized in 1999 when Günter Blobel received the Nobel Prize in physiology or medicine for his discovery "proteins have intrinsic signal that govern their transport and localization in the cell". The press release can be read here.

— Was SignalP the first signal peptide predictor?

No, but it was, to our knowledge, the first to be implemented as a web server (in 1996). Among the earlier methods were McGeoch (1985) and von Heijne (1986), both of which have been included in PSORT.

— How many times have the SignalP papers been cited?

This information is available on Henrik Nielsen's ResearcherID, Scopus, and Google Scholar pages.

References

---

Main references:

• Original method (SignalP v. 1.1)
• Update to SignalP v. 2.0
• Update to SignalP v. 3.0
• Update to SignalP v. 4.0
• Update to SignalP v. 4.1 (current method)

Other publications
Henrik Nielsen's PhD thesis

---

Original method (SignalP v. 1.1)

Identification of prokaryotic and eukaryotic signal peptides and prediction of their cleavage sites.
Henrik Nielsen, Jacob Engelbrecht, Søren Brunak and Gunnar von Heijne.
Protein Engineering, 10:1-6 (1997).

We have developed a new method for the identification of signal peptides and their cleavage sites based on neural networks trained on separate sets of prokaryotic and eukaryotic sequence. The method performs significantly better than previous prediction schemes and can easily be applied on genome-wide data sets. Discrimination between cleaved signal peptides and uncleaved N-terminal signal-anchor sequences is also possible, though with lower precision. Predictions can be made on a publicly available WWW server.

PMID: 9051728 (free full text pdf version)

---

Update to SignalP v. 2.0

Prediction of signal peptides and signal anchors by a hidden Markov model.
Henrik Nielsen and Anders Krogh.
Proc Int Conf Intell Syst Mol Biol. (ISMB 6), 6:122-130 (1998).

A hidden Markov model of signal peptides has been developed. It contains submodels for the N-terminal part, the hydrophobic region, and the region around the cleavage site. For known signal peptides, the model can be used to assign objective boundaries between these three regions. Applied to our data, the length distributions for the three regions are significantly different from expectations. For instance, the assigned hydrophobic region is between 8 and 12 residues long in almost all eukaryotic signal peptides. This analysis also makes obvious the difference between eukaryotes, Gram-positive bacteria, and Gram-negative bacteria. The model can be used to predict the location of the cleavage site, which it finds correctly in nearly 70% of signal peptides in a cross-validated test — almost the same accuracy as the best previous method. One of the problems for existing prediction methods is the poor discrimination between signal peptides and uncleaved signal anchors, but this is substantially improved by the hidden Markov model when expanding it with a very simple signal anchor model.

PMID: 9783217

---

Update to SignalP v. 3.0

Improved prediction of signal peptides: SignalP 3.0.
Jannick Dyrløv Bendtsen, Henrik Nielsen, Gunnar von Heijne and Søren Brunak.
J. Mol. Biol., 340:783-795 (2004).

We describe improvements of the currently most popular method for prediction of classically secreted proteins, SignalP. SignalP consists of two different predictors based on neural network and hidden Markov model algorithms, and both components have been updated. Motivated by the idea that the cleavage site position and the amino acid composition of the signal peptide are correlated, new features have been included as input to the neural network. This addition, together with a thorough error-correction of a new data set, have improved the performance of the predictor significantly over SignalP version 2. In version 3, correctness of the cleavage site predictions have increased notably for all three organism groups, eukaryotes, Gram negative and Gram positive bacteria. The accuracy of cleavage site prediction has increased in the range from 6–17 % over the previous version, whereas the signal peptide discrimination improvement mainly is due to the elimination of false positive predictions, as well as the introduction of a new discrimination score for the neural network. The new method has also been benchmarked against other available methods.

PMID: 15223320         doi: 10.1016/j.jmb.2004.05.028

---

Update to SignalP v. 4.0

SignalP 4.0: discriminating signal peptides from transmembrane regions.
Thomas Nordahl Petersen, Søren Brunak, Gunnar von Heijne and Henrik Nielsen.
Nature Methods, 8:785-786 (2011).

This is a Correspondence, it has no abstract.

doi: 10.1038/nmeth.1701
Access to the paper: if you have a personal or institutional subscription to Nature Methods, use the doi: link above. Access to the supplementary materials: nmeth.1701-S1.pdf

---

Other publications

Locating proteins in the cell using TargetP, SignalP, and related tools
Olof Emanuelsson, Søren Brunak, Gunnar von Heijne, Henrik Nielsen
Nature Protocols, 2:953-971 (2007).

Determining the subcellular localization of a protein is an important first step toward understanding its function. Here, we describe the properties of three well-known N-terminal sequence motifs directing proteins to the secretory pathway, mitochondria and chloroplasts, and sketch a brief history of methods to predict subcellular localization based on these sorting signals and other sequence properties. We then outline how to use a number of internet-accessible tools to arrive at a reliable subcellular localization prediction for eukaryotic and prokaryotic proteins. In particular, we provide detailed step-by-step instructions for the coupled use of the amino-acid sequence-based predictors TargetP, SignalP, ChloroP and TMHMM, which are all hosted at the Center for Biological Sequence Analysis, Technical University of Denmark. In addition, we describe and provide web references to other useful subcellular localization predictors. Finally, we discuss predictive performance measures in general and the performance of TargetP and SignalP in particular.

PMID: 17446895
Please click here to access the paper and supplementary materials.

Machine learning approaches to the prediction of signal peptides and other protein sorting signals.
Henrik Nielsen, Søren Brunak, and Gunnar von Heijne.
Protein Engineering, 12:3-9 (1999), Review.

Prediction of protein sorting signals from the sequence of amino acids has great importance in the field of proteomics today. Recently, the growth of protein databases, combined with machine learning approaches, such as neural networks and hidden Markov models, have made it possible to achieve a level of reliability where practical use in, for example automatic database annotation is feasible. In this review, we concentrate on the present status and future perspectives of SignalP, our neural network-based method for prediction of the most well-known sorting signal: the secretory signal peptide. We discuss the problems associated with the use of SignalP on genomic sequences, showing that signal peptide prediction will improve further if integrated with predictions of start codons and transmembrane helices. As a step towards this goal, a hidden Markov model version of SignalP has been developed, making it possible to discriminate between cleaved signal peptides and uncleaved signal anchors. Furthermore, we show how SignalP can be used to characterize putative signal peptides from an archaeon, Methanococcus jannaschii. Finally, we briefly review a few methods for predicting other protein sorting signals and discuss the future of protein sorting prediction in general.

PMID: 10065704

A neural network method for identification of prokaryotic and eukaryotic signal peptides and prediction of their cleavage sites.
Henrik Nielsen, Jacob Engelbrecht, Søren Brunak and Gunnar von Heijne.
Int. J. Neural Sys., 8:581-599 (1997).

We have developed a new method for the identification of signal peptides and their cleavage sites based on neural networks trained on separate sets of prokaryotic and eukaryotic sequences. The method performs significantly better than previous prediction schemes, and can easily be applied to genome-wide data sets. Discrimination between cleaved signal peptides and uncleaved N-terminal signal-anchor sequences is also possible, though with lower precision.

PMID: 10065837

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, 24:165-77 (1996).

When preparing data sets of amino acid or nucleotide sequences it is necessary to exclude redundant or homologous sequences in order to avoid overestimating the predictive performance of an algorithm. For some time methods for doing this have been available in the area of protein structure prediction. We have developed a similar procedure based on pair-wise alignments for sequences with functional sites. We show how a correlation coefficient between sequence similarity and functional homology can be used to compare the efficiency of different similarity measures and choose a nonarbitrary threshold value for excluding redundant sequences. The impact of the choice of scoring matrix used in the alignments is examined. We demonstrate that the parameter determining the quality of the correlation is the relative entropy of the matrix, rather than the assumed (PAM or identity) substitution mode. Results are presented for the case of prediction of cleavage sites in signal peptides. By inspection of the false positives, several errors in the database were found. The procedure presented may be used as a general outline for finding a problem-specific similarity measure and threshold value for analysis of other functional amino acid or nucleotide sequence patterns.

PMID: 8820484

From sequence to sorting: Prediction of signal peptides.
Henrik Nielsen.
Ph.D. thesis, defended at Department of Biochemistry, Stockholm University, Sweden, May 25, 1999.

In the present age of genome sequencing, a vast number of predicted genes are initially known only by their putative nucleotide sequence. The newly established field of bioinformatics is concerned with the computational prediction of structural and functional properties of genes and the proteins they encode, based on their nucleotide and amino acid sequences.
      Since one of the crucial properties of a protein is its subcellular location, prediction of protein sorting is an important question in bioinformatics. A fundamental distinction in protein sorting is that between secretory and non-secretory proteins, determined by a cleavable N-terminal sorting signal, the secretory signal peptide.
      The main part of this thesis, including four of the six papers, concerns prediction of secretory signal peptides in both eukaryotic and bacterial data using two machine learning techniques: artificial neural networks and hidden Markov models. A central result is the SignalP prediction method, which has been made available as a World Wide Web server and is very widely used.
      Two additional prediction methods are also included, with one paper each. ChloroP predicts chloroplast transit peptides, another cleavable N-terminal sorting signal; while NetStart predicts start codons in eukaryotic genes. For prediction of all N-terminal signals, the assignment of correct start codon can be critical, which is why prediction of translation initiation from the nucleotide sequence is also important for protein sorting prediction.
      This thesis comprises a detailed review of the molecular biology of protein secretion, a short introduction to the most important machine learning algorithms in bioinformatics, and a critical review of existing methods for protein sorting prediction. In addition, it contains general treatment of the principles of data set construction and performance evaluation for prediction methods in bioinformatics.

Access to the thesis (without the six included papers): PhDthesis.pdf; PhDthesis-cover.pdf

Instructions

---

Output format

---

DESCRIPTION OF THE SCORES

The neural networks in SignalP produce three output scores for each position in the input sequence:

C-score (raw cleavage site score)

The output from the CS networks, which are trained to distinguish signal peptide cleavage sites from everything else.
Note the position numbering of the cleavage site: the C-score is trained to be high at the position immediately after the cleavage site (the first residue in the mature protein).

S-score (signal peptide score)

The output from the SP networks, which are trained to distinguish positions within signal peptides from positions in the mature part of the proteins and from proteins without signal peptides.

Y-score (combined cleavage site score)

A combination (geometric average) of the C-score and the slope of the S-score, resulting in a better cleavage site prediction than the raw C-score alone. This is due to the fact that multiple high-peaking C-scores can be found in one sequence, where only one is the true cleavage site. The Y-score distinguishes between C-score peaks by choosing the one where the slope of the S-score is steep.

The graphical output from SignalP (see below) shows the three different scores, C, S and Y, for each position in the sequence.

In the summary below the plot, the maximal values of the three scores are reported. In addition, the following two scores are shown:

mean S

The average S-score of the possible signal peptide (from position 1 to the position immediately before the maximal Y-score).

D-score (discrimination score)

A weighted average of the mean S and the max. Y scores. This is the score that is used to discriminate signal peptides from non-signal peptides.

For non-secretory proteins all the scores represented in the SignalP output should ideally be very low (close to the negative target value of 0.1).

EXAMPLE OUTPUT

By default the server produces the following output for each input sequence:

Example: secretory protein — standard output format

The example below shows the output for thioredoxin domain containing protein 4 precursor (endoplasmic reticulum protein ERp44), taken from the Swiss-Prot entry ERP44_HUMAN. The signal peptide prediction is consistent with the database annotation.

# SignalP-4.1 euk predictions
>sp_Q9BS26_ERP44_HUMAN  Endoplasmic reticulum resident protein 44 OS_Homo sapiens GN_ERP44 PE_1 SV_1


# Measure  Position  Value    Cutoff   signal peptide?
  max. C    30       0.427
  max. Y    30       0.586
  max. S     9       0.950
  mean S     1-29    0.821
       D     1-29    0.713   0.450   YES
Name=sp_Q9BS26_ERP44_HUMAN	SP='YES' Cleavage site between pos. 29 and 30: VTT-EI D=0.713 D-cutoff=0.450 Networks=SignalP-noTM
# data
# gnuplot script

Signal peptides: 1
# processed fasta entries
# gff file of processed entries

Below the summary for each sequence, two files are provided via links: "data" and "gnuplot script". If you have the free graphics program gnuplot on your computer, you can use these two files to customize your plot.

Below the output for all the sequences, two other files are provided via links, if at least one signal peptide has been predicted. These are "processed fasta entries", a FASTA sequence file containing the sequences of protein that had predicted signal peptides, with the signal peptide removed; and "gff file of processed entries", a file showing the signal peptides feature of those proteins that had predicted signal peptides in GFF format.

Example: secretory protein — short output format

# SignalP-4.0 euk predictions
# name                     Cmax  pos  Ymax  pos  Smax  pos  Smean   D     ?  Dmaxcut    Networks-used
ERP44_HUMAN                0.427  30  0.586  30  0.950   9  0.821   0.713 Y  0.450      SignalP-noTM

Performance of SignalP 4.1

---

Correlation

In the SignalP 4.0 article, we show that SignalP 4.0 is superior in performance to SignalP 3.0 and ten competing methods (five dedicated signal peptide predictors and five transmembrane topology predictors with built-in signal peptide models), when the performance is measured by Matthews Correlation Coefficient (MCC).

Matthews Correlation Coefficient is a very widely used measure for performance in bioinformatics. It is defined thus:

${\displaystyle \mathrm{MCC}=\frac{\mathrm{tp}\times \mathrm{tn}-\mathrm{fp}\times \mathrm{fn}}{\sqrt{\left(\mathrm{tp}+\mathrm{fp}\right)\left(\mathrm{tp}+\mathrm{fn}\right)\left(\mathrm{tn}+\mathrm{fp}\right)\left(\mathrm{tn}+\mathrm{fn}\right)}}}$ where

• tp is the number of true positives (signal peptides predicted as such)
• tn is the number of true negatives (non-signal peptides predicted as such)
• fp is the number of false positives (erroneous signal peptide predictions)
• fn is the number of false negatives (missed signal peptides)

and it takes the value of 1 for a perfect prediction, 0 for a random (non-informative) prediction, and -1 for a consistently wrong prediction.

In Table E (pp. 10-11) of the supplementary materials you can see the MCC values for SignalP and the competing methods.

Sensitivity, false positive rate and cutoff choice

However, SignalP 4.0 is not superior to SignalP 3.0 according to all performance measures. Notably, the sensitivity is lower when you use the default cutoff. Sensitivity is the proportion of the true signal peptides that are correctly predicted:

${\displaystyle \mathrm{Sens}=\frac{\mathrm{tp}}{\mathrm{tp}+\mathrm{fn}}}$

All prediction methods that make a classification from a numerical output have a choice to make: where to place the cutoff (also known as threshold) for the output? If you use a high cutoff, you will get few false positives, but also a low sensitivity; if you lower the cutoff, you will get a better sensitivity at the price of more false positives. The false positive rate is defined as:

${\displaystyle \mathrm{FPR}=\frac{\mathrm{fp}}{\mathrm{fp}+\mathrm{tn}}}$

There is no single correct answer to the problem of choosing the cutoff, it depends on the contet in which the prediction method is used. For SignalP, we have used a cutoff on the D-score (see the Output format for a definition) that maximizes the MCC.

ROC curves

The trade-off between sensitivity and false positive rate is often illustrated graphically as a so-called ROC curve which has false positive rate on the x-axis and sensitivity on the y-axis for varying values of the cutoff. The better a predition method is, the closer to the upper left corner the ROC curve will be, while a random (non-informative) prediction will follow the diagonal. This is an excellent way to compare different predictors, since it is not dependent on cutoff choice.

Below, you can see ROC curves for SignalP 3 and 4 for the three different organism groups. Note: in contrast to the values in Table E, these are not evaluation performances; they are made by applying the finished methods to the Total data set before homology reduction.

These ROC curves show that:

• When there are TM segments in the data ("all data"), SignalP 4.0 is clearly better than SignalP 3.0 (compare the pink and green curves)
• When TM segments are excluded from the data ("no TM"), SignalP 4.0 performance is practically equal to that of SignalP 3.0 — except in the Gram-positives, where it is better (compare the blue and red curves)
• SignalP 4.0 and 3.0 default cutoffs are placed at very different points on the ROC curves, leading to lower sensitivity (and much lower FP rates) in SignalP 4.0.

The cutoff choice in SignalP 4.1

SignalP 4.1 offers the users an option of using cutoff values which reproduce the sensitivity of SignalP 3.0. The price is, of course, a slightly higher false positive rate.

In the table below, the performace values are shown for SignalP 3.0, SignalP 4.1 with default cutoff, and SignalP 4.1 with "sensitive" (SignalP-3.0 compliant) cutoff. Note, again, that these are not evaluation performances and should not be used to compare SignalP to competing methods, they are merely for the purpose of comparing SignalP versions.

Method	Cutoff, SignalP-noTM	Cutoff, SignalP-TM	Sensitivity	FP rate, no TM	FP rate, all data	MCC, no TM	MCC, all data
Eukaryotic data
SignalP 3.0	0.43		0.988	0.008	0.117	0.978	0.781
SignalP 4.1 default	0.45	0.50	0.967	0.003	0.011	0.972	0.955
SignalP 4.1 sensitive	0.34	0.34	0.988	0.009	0.043	0.976	0.903
Gram-positive data
SignalP 3.0	0.45		0.961	0.008	0.033	0.937	0.814
SignalP 4.1 default	0.57	0.45	0.950	0.000	0.001	0.973	0.967
SignalP 4.1 sensitive	0.42	0.42	0.961	0.000	0.003	0.978	0.958
Gram-negative data
SignalP 3.0	0.44		0.955	0.004	0.061	0.949	0.691
SignalP 4.1 default	0.57	0.51	0.924	0.000	0.001	0.957	0.949
SignalP 4.1 sensitive	0.42	0.42	0.955	0.002	0.006	0.963	0.937

Examples of proteome predictions for three organism types

---

Eukaryota - Human proteom GRCh37.62

short
gff
mature

Gram positive bacteria - B.subtilis EB2

short
gff
mature

Gram negative bacteria - E.coli K12

short
gff
mature

---

Training and testing data sets

These are the annotated sequence data described in Table A of the Supplementary Materials. The entire datasets correspond to the "Total" columns in the table (before homology reduction). Sequences labeled "Train" correspond to the "Train" columns in the table, while sequences labeled "Evaluation" correspond to the "Comp." columns in the table (used for comparing the performance to SignalP 3.0 and other methods). Sequences used to train SignalP 3.0 (or homologous to those used to train SignalP 3.0) have been removed from the "Comp." sets.

Note that the "Comp." sets are subsets of the "Train" sets. The evaluation of SignalP 4.0 was done using a nested cross-validation approach, where different partitions were used for training, optimization and evaluation, see Supplementary Materials for details.

   166 AJL2_ANGJA Evaluation
MVSFKLPAFLCVAVLSSMALVSHGAVLGLCEGACPEGWVEHKNRCYLHVAEKKTWLDAELNCLHHGGNLASEHSEDEHQF
LKDLHKGSDDPFWIGLSAVHEGRSWLWSDGTSASAEGDFSMWNPGEPNDAGGKEDCVHDNYGGQKHWNDIKCDLLFPSIC
VLRMVE
SSSSSSSSSSSSSSSSSSSSSSSS........................................................
................................................................................
......
   503 A1BG_BOVIN  Evaluation Train
MSAWAALLLLWGLSLSPVTEQATFFDPRPSLWAEAGSPLAPWADVTLTCQSPLPTQEFQLLKDGVGQEPVHLESPAHEHR
FPLGPVTSTTRGLYRCSYKGNNDWISPSNLVEVTGAEPLPAPSISTSPVSWITPGLNTTLLCLSGLRGVTFLLRLEGEDQ
FLEVAEAPEATQATFPVHRAGNYSCSYRTHAAGTPSEPSATVTIEELDPPPAPTLTVDRESAKVLRPGSSASLTCVAPLS
GVDFQLRRGAEEQLVPRASTSPDRVFFRLSALAAGDGSGYTCRYRLRSELAAWSRDSAPAELVLSDGTLPAPELSAEPAI
LSPTPGALVQLRCRAPRAGVRFALVRKDAGGRQVQRVLSPAGPEAQFELRGVSAVDSGNYSCVYVDTSPPFAGSKPSATL
ELRVDGPLPRPQLRALWTGALTPGRDAVLRCEAEVPDVSFLLLRAGEEEPLAVAWSTHGPADLVLTSVGPQHAGTYSCRY
RTGGPRSLLSELSDPVELRVAGS
SSSSSSSSSSSSSSSSSSSSS...........................................................
................................................................................
................................................................................
................................................................................
................................................................................
................................................................................
.......................

The format is:

1. First a header line with number of amino acids, sequence name (UniProt ID) and possibly a description field ('Evaluation'/'Train').
2. The protein sequence.
3. The annotations, one for each amino acid.

Annotations:
S — Amino acid is part of a Signal peptide (experimentally verified)
T — Amino acid is part of a Transmembrane region (experimentally verified)
t — Amino acid is part of a Transmembrane region (not experimentally verified)
. — An annotation different from those shown above

Eukaryota sequence data
Gram positive sequence data
Gram negative sequence data

Version history

---

Please click on the version number to activate the corresponding server where available.

4.1	The current server. New in this version: For the web page, an option to set the D-score cutoff values so that the sensitivity is the same as that of SignalP 3.0. Option included to set the minimum cleavage site position i.e. Ymax position - default value is 10. For the signalp package an option has been included to specify a temporary directory (-T dir). For the signalp package an option has been included to show signalp version (-V). Documentation rewritten. Main publication: SignalP 4.0: discriminating signal peptides from transmembrane regions Thomas Nordahl Petersen, Søren Brunak, Gunnar von Heijne and Henrik Nielsen. Nature Methods, 8:785-786, 2011.
4.0	New in this version: Improved discrimination between signal peptides and transmembrane regions. No HMM method - only one prediction. Main publication: SignalP 4.0: discriminating signal peptides from transmembrane regions Thomas Nordahl Petersen, Søren Brunak, Gunnar von Heijne and Henrik Nielsen. Nature Methods, 8:785-786, 2011.
3.0	New in this version: D-score. Improved quality of prediction. Main publication: Improved prediction of signal peptides: SignalP 3.0. Jannick Dyrløv Bendtsen, Henrik Nielsen, Gunnar von Heijne and Søren Brunak. J. Mol. Biol., 340:783-795, 2004.
2.0	New in this version: Incorporation of a hidden Markov model version: SignalP V2.0 comprises two signal peptide prediction methods, SignalP-NN (based on neural networks, corresponding to SignalP V1.1) and SignalP-HMM (based on hidden Markov models). For eukaryotic data, SignalP-HMM has a substantially improved discrimination between signal peptides and uncleaved signal anchors, but it has a slightly lower accuracy in predicting the precise location of the cleavage site. The user can choose whether to run SignalP-NN, SignalP-HMM, or both. Retraining of the neural networks: SignalP-NN in SignalP V2.0 is trained on a newer data set derived from SWISS-PROT rel. 35 (instead of rel. 29 as in SignalP V1.1). Graphics integrated in the output: SignalP V2.0 shows signal peptide and cleavage site scores for each position as plots in GIF format on the output page. The plots provide more information than the prediction summary, e.g. about possible cleavage sites other than the strongest prediction. Signal peptide region assignment: SignalP-HMM provides not only a prediction of the presence of a signal peptide and the position of the cleavage site, but also an approximate assignment of n-, h- and c-regions within the signal peptide. These are shown in the graphical output as probabilities for each position being in one of these three regions. Automatic truncation: in SignalP V1.1, we recommended that you should submit only the N-terminal part of each protein, not more than 50-70 amino acids. SignalP V2.0 now offers to truncate your sequences automatically. Main publication: Prediction of signal peptides and signal anchors by a hidden Markov model. Henrik Nielsen and Anders Krogh. Proceedings of the Sixth International Conference on Intelligent Systems for Molecular Biology (ISMB 6), AAAI Press, Menlo Park, California, pp. 122-130, 1998.
1.1	The original server: the method based on artificial neural networks. Main publication: Identification of prokaryotic and eukaryotic signal peptides and prediction of their cleavage sites. Henrik Nielsen, Jacob Engelbrecht, Søren Brunak and Gunnar von Heijne. Protein Engineering, 10:1-6, 1997.

Software Downloads

---

• Version 6.0h
• slow_sequential
• fast
• Version 5.0b
• Linux
• Darwin
• Version 4.1g
• Linux
• IRIX64
• Darwin
• CYGWIN
• Version 3.0
• SunOS
• OSF1
• Linux
• IRIX
• AIX
• Version 2.0
• SunOS
• OSF1
• Linux
• IRIX
• AIX