Task 3: Sequence-based predictions

Task description

The full description of this task can be found here.

Task 3.3: Prediction of transmembrane alpha-helices and signal peptides

Annotated sequence features

PAH

The phenylalanine-4-hydroxylase has no annotated signal peptide or transmembrane helices.

BACR_HALSA

The bacteriorhodopsin has the following annotated signal peptide and transmembrane helices:

Position	Feature Name	Description
1 - 13	Propeptide
14 – 23	Topological domain	Extracellular
24 - 42	Transmembrane	Helical; Name=Helix A
43 – 56	Topological domain	Cytoplasmic
57 - 75	Transmembrane	Helical; Name=Helix B
76 – 91	Topological domain	Extracellular
92 - 109	Transmembrane	Helical; Name=Helix C
110 – 120	Topological domain	Cytoplasmic
121 - 140	Transmembrane	Helical; Name=Helix D
141 – 147	Topological domain	Extracellular
148 - 167	Transmembrane	Helical; Name=Helix E
168 – 185	Topological domain	Cytoplasmic
186 - 204	Transmembrane	Helical; Name=Helix F
205 – 216	Topological domain	Extracellular
217 - 236	Transmembrane	Helical; Name=Helix G
237 – 262	Topological domain	Cytoplasmic

RET4_HUMAN

The retinol-binding protein 4 has the following annotated signal peptide (no transmembrane helices are annotated):

Position	Feature Name	Description
1 - 18	Signal peptide

INSL5_HUMAN

The Insulin-like peptide INSL5 has the following annotated signal peptide (no transmembrane helices are annotated):

Position	Feature Name	Description
1 - 22	Signal peptide

LAMP1_HUMAN

The lysosome-associated membrane glycoprotein 1 has the following annotated signal peptide and transmembrane helices:

Position	Feature Name	Description
1 - 28	Signal peptide
29 – 382	Topological domain	Lumenal
383 - 405	Transmembrane	Helical;
406 – 417	Topological domain	Cytoplasmic

A4_HUMAN

The Amyloid beta A4 protein has the following annotated signal peptide and transmembrane helices:

Position	Feature Name	Description
1 - 17	Signal peptide
18 – 699	Topological domain	Extracellular
700 - 723	Transmembrane	Helical;
724 – 770	Topological domain	Cytoplasmic

General Questions to prediction of transmembrane alpha-helices and signal peptides

Why is the prediction of transmembrane helices and signal peptides grouped together here?

Methods which only predict transmembrane helices often predict signal peptides as transmembrane helices as well. The reason for this is that both, transmembrane helices and signal peptides consist mainly of hydrophobic residues. These false predictions lead to inaccurate topological features and thus to wrongly annotated function of a protein. To avoid these cases most recent methods couple their transmembrane prediction together with a signal peptide prediction.

Description of different signal peptides

Signalpeptides for the import to the endoplasmic reticulum (ER)

The import to the ER is usually required for the secretory pathway (to export proteins out of a cell). The import process can occur either co-translational (the nascent protein chain is translocated together with the ribosome) or post-translational (only the fully synthesized protein is transported to the ER). However, for both cases the SEC-pathway is mostly used.

The co-translational transport to the ER is done by the signal recognition particle (SRP). This particle recognizes the N-terminal signal-sequence of the nascent polypeptide chain and then transports it to the ER membrane where the complex, consisting of SRP, polypeptide chain and ribosome, is recognized by the ER membrane bound signal recognition particle receptor (SR). After this recognition the polypeptide chain is imported into the ER lumen via the SEC channel in an ATP dependent process.

The post-translational import to the ER lumen is done by chaperons which guide the polypeptide chain to the SEC channel which is then imported in an ATP dependent process.

However, not only the import to the ER lumen is possible, an import to the ER membrane is possible as well. So far, 5 different types of import to the ER membrane are known.

Type 1 requires an N-terminal signal sequence and an intrinsic stop transfer anchor sequence which will be the part which is inserted in the membrane.

Type 2 and 3 do not require a N-terminal signal sequence only a intrinsic signal anchor sequence is required. The difference between type 2 and 3 is that type 2 has positively charged residues before the signal anchor sequence (on the N-Terminal side) and type 3 has positively charged residues after the signal anchor sequence (on C-Terminal side). These charged residues of trans-membrane protein are always in the cytosol. Thus, type 2 inserted proteins have their N-terminal end residing in the cytosol whereas type 3 inserted proteins have a C-terminal end in the cytosol.

Type 4-A and 4-B insertion is also known as multipass membrane insertion. These proteins have not one trans-membrane helix like the proteins imported via Type 1,2 and 3, instead they have several trans-membrane helices. Hence, they consist of multiple internal stop-transfer anchor sequences and internal signal-anchor sequences. The difference between type 4-A and 4-B is that in type 4-A the N and C terminal ends are located in the cytosol whereas type 4-B import results in a N-terminal end residing in the ER lumen and a C-terminal end residing in the cytosol.

In addition to the N-terminal import of trans-membrane proteins there is also the possiblity for a C-terminal import. Obviously, these proteins are imported post-translation.

Signalpeptides for the import to the mitochondrion

There are several targets for import to the mitochondrion, proteins can be translocated to the matrix, the outer membrane, the inner membrane and the inter membrane space.

Proteins who are designated to be imported to the matrix of a mitochondrion have a N-terminal matrix-targeting sequence. This mitochondrial import to the matrix is assisted by chaperons (Hsc70) which guide the protein to the import pore complex of the mitochondrion. The import through the outer membrane is conducted by the TOM complex and the following import through the inner membrane is conducted by the TOM complex. After successful import to the matrix the signal sequence is cleaved off by proteolytic active enzymes.

Import to the inner membrane can occur in three ways. The first way is the TIM22 pathway, proteins using this pathway need internal targeting sequences. The next way is the stop transfer import, for this proteins need a stop transfer sequence and a N-terminal matrix targeting sequence. The third way is called conservative sorting proteins using this pathway have a N-terminal targeting sequence as well and in addition intrinsic Oxa1-targeting sequences which are recognized by Ox1-proteins which execute the import to the membrane.

Proteins imported to the outer membrane of a mitochondrion usually have PORTA domains which are recognized by the TOB/SAM complex.

Signalpeptides for the import to the chloroplast

Proteins heading to chloroplasts can target different parts of it. For example the stroma, inner and outer membrane, the thylakoids membrane or the thylakoids lumen.

Usually these protein have a N-terminal targeting sequence.

Signalpeptides for the import to the peroxisome

Peroxisomal proteins can be imported to the lumen or to the membrane. Proteins imported to the lumen have either a peroxisomal targeting signal at the C-termins (also known as PTS1) or a targeting sequence close to the N-terminus (also known as PTS2). Proteins imported to the membrane can have an intrinsic membrane peroxisomal targeting signal (mPTS). However, not all proteins have this mPTS. These proteins are imported to the ER and from there they bud off together with the mature peroxisome.

Signalpeptides for the import to the nucleus and the export form the nucleus

Proteins which are imported to the nucleus require a nuclear localisation signal (NLS) which is recognized by importin. The NLS containing protein is then imported via the nuclear pore complex (NPC) to the nucleoplasm.

Proteins which are exported from the nucleus require a nuclear export signal which is recognized by exportin, a protein which binds to the NES of the cargo protein. In addition to exportin a second component, known as Ran*GTP, is required to mediate the export through the NPC.

TMHMM

Details of the method

Author: Sonnhammer, Heijne & Krogh

Year: 1998

Reference: PubMed

Description

HMM architecture of TMHMM Disclaimer: This file is redistributed from [Krogh A, Larsson B, von Heijne G, Sonnhammer EL. Predicting transmembrane protein topology with a hidden Markov model: application to complete genomes. J Mol Biol. 2001 Jan 19;305(3):567-80.] . All rights belong to the creator.

This method is based on a hidden markov model (HMM). The authors of this method tried to model the 'grammar' of transmembrane proteins in order to predict the protein topology of transmembrane more accurate than methods who only e.g. rely on propensity values and do not consider the topological constraints of these class of proteins.

TMHMM defined for their HMM for each feature one or more states which present this feature. For example the transmembrane helix is modeled by three sub models. A model for the helix core, the cap of the helix which lies partly in the cytoplasm and the membrane and the cap which is partly in the membrane and cytoplasm. In addition to this helix model they also created sub models for the cytoplasmic loop and the non-cytoplasmic loop as well as a sub model for the globular region. Each sub model can reflect one or more states in the HMM model. For example the globular sub model only consists of one HMM state whereas the helix-core and caps are modeled by multiple HMM states.

The 'grammar' is incorporated to this HMM model by defining the possible transitions from one sub model to another one. For example it is only possible to change from a cytoplasmic loop region to a cytoplasmic cap region and then to the helix core and after that either to non-cytoplasmic short loop or long non-cytoplasmic loop and so on.

Predicted features

This methods predicts the transmembrane helix and whether this part is in the cytoplasm (in) or outside of it (out).

Required information for the prediction

User who want to use it just need their amino acid sequence of their query sequence. The transmission and emission probabilities are derived from 160 transmembrane protein sequences.

Phobius

Details of the method

Author: Käll, Krogh, Sonnhammer

Year: 2004

Reference: PubMed

Description

HMM architecture of Phobius Disclaimer: This file is redistributed from [KKäll L, Krogh A, Sonnhammer EL. A combined transmembrane topology and signal peptide prediction method. J Mol Biol. 2004 May 14;338(5):1027-36.] . All rights belong to the creator.

Phobius is an HMM based prediction method to predict transmembrane helices as well as N-terminal signal peptides. More precisely, it is a combination of the two HMM models of TMHMM and SignalP which is merged into one HMM. This was done in order to overcome problems associated with transmembrane helix prediction: signale peptides are often wrongly predicted as transmembrane helices. The complete architecture can be seen in the figure.

Predicted features

Phobius predicts transmembrane helices, signal peptides and the topology of the loops (whether they are inside the cytoplasm or not).

Required information for the prediction

Users only has to enter the amino acid sequence of their query protein in FASTA format.

Execution

Results and discussion

PolyPhobius

Details of the method

Author: Käll L, Krogh A, Sonnhammer EL

Year: 2005

Reference: PubMed

Description

PolyPhobius is also based on a HMM which constraints the possible transitions from one state to another in order to reflect the 'grammar' of transmembrane proteins. However, the difference to the ordinary Phobius is that it uses knowledge homologous sequences of the query sequences as well to make the prediction more accurate.

In order to do so it calculates for each sequence position for each label (e.g. transmembrane helix, in, out, etc...) for each homologous sequence the posterior label probability (PLP). The PLP is defined as "the probability of a label at a certain position in the sequence, given the sequence and the model" (quoted from "Käll L, Krogh A, Sonnhammer EL. An HMM posterior decoder for sequence feature prediction that includes homology information Bioinformatics. 2005 Jun;21 Suppl 1:i251-7."). Then a multiple sequence alignment (MSA) of all homologous sequences is build, for each position in the MSA a average PLP is calculated. This average PLP will be then be used by the optimal accuracy algorithm to predict the most likely sequences of states for a given query sequence and thus the topology of the transmembrane helices.

Predicted features

This method predicts the same features as the ordinary Phobius, which means transmembrane helices, the signal peptide and whether the connecting loops of transmembrane helices are inside or outside.

Required information for the prediction

User need the amino acid sequence of their protein in FASTA format. An additional option is to specify the homologous sequences manually. If that is not done PolyPhobius will search for homologous sequences by itself by using BLAST.

OCTOPUS

Details of the method

Author: Viklund H, Elofsson A.

Year: 2008

Reference: Bioinformatics

Description

Flowchart of OCTOPUS Disclaimer: This file is redistributed from [Viklund H, Elofsson A. OCTOPUS: improving topology prediction by two-track ANN-based preference scores and an extended topological grammar. Bioinformatics. 2008 Aug 1;24(15):1662-8. Epub 2008 May 12.] . All rights belong to the creator.

OCTOPUS basically uses two methods to predict the topology of transmembrane proteins: artificial neural networks (ANN) and hidden markov models (HMM). In a first step BLAST searches for homologous sequences of a input FASTA sequence. From the found homologous sequences a multiple sequence alignment is build from which a raw sequence profile and a sequence profile based on PSSM are extracted. These profiles are used for two sets of ANNs.

The first set of ANNs contains four separate ANNs which predict the residue preference for M (Membrane), I (Interface), L (Loop), G (Globular). In order to make the predictions for G and M more smooth the output of the first row of ANNs output is used for a second ANN as input.The second set of ANNs is taken to predict the residue preferences for the inside/outside residues.

Finally. the output of these two sets of ANNs are used to parameterize the OCTOPUS-HMM for the actual topological feature prediction. This HMM is needed to model the 'grammar' of trans membrane proteins, which simply means that only certain state transitions are allowed. For example, if we assume we are currently in the transmembrane state then it is only allowed to go into the loop state and so on and so forth.

The state sequences which fits best the input sequence is then calculated by the Viterbi algorithm.

Predicted features

Predicted features are inside/outside (i/o), transmembrane (M), TM hairpin (H), reentrant (R) or membrane dip (D)

Required information for the prediction

Only the amino acid sequence of the users protein is required.

SPOCTOPUS

Details of the method

Author: Viklund H, Bernsel A, Skwark M, Elofsson A.

Year: 2008

Reference: Bioinformatics

Description

SPOCTOPUS works the same way as OCTOPUS does. The only difference is that it includes a signal peptide prediction.

Predicted features

Predicted features are signal peptide, inside/outside (i/o), transmembrane (M), TM hairpin (H), reentrant (R) or membrane dip (D)

Required information for the prediction

Only the amino acid sequence of the query protein is required as input.

SignalP

Details of the method

Author: Henrik Nielsen, Jacob Engelbrecht, Søren Brunak and Gunnar von Heijne.

Year: 1997

Reference: PubMed

Description

This predictor takes two methods into account the first method used is a neural network the second is a hidden markov model.

There are two neural networks one which is predicting whether the first n amino acids belong to a signal peptide and the second network predicts the exact cleavage side positon.

In a later version of SignalP a hidden markov model (HMM) has been also build to predict signal peptides. However, this prediction is completely independent from the neural network prediction. This HMM models the N-terminal region of a signal peptide as well as the surrounding cleavage site.

Predicted features

Predicts the presence of signal peptidase I cleavage sites and whether the first n residues belong to a signal peptide.

Required information for the prediction

The amino acid sequence of the protein and whether this protein is from a eukaryote, gram-negative bacteria or gram-positive bacteria.

TargetP

Details of the method

Author: Henrik Nielsen, Jacob Engelbrecht, Søren Brunak and Gunnar von Heijne.

Year: 1997

Reference: PubMed

Description

Architecture of TargetP Disclaimer: This file is redistributed from Emanuelsson O, Nielsen H, Brunak S, von Heijne G. Predicting subcellular localization of proteins based on their N-terminal amino acid sequence. J Mol Biol. 2000 Jul 21;300(4):1005-16.] . All rights belong to the creator.

TargetP's prediction are based on trained neural networks. These neural networks are build up in a two layer setup. The first layer consists of three neural networks which are used to predict whether it is a chloroplast targeting sequence, a mitochondrial targeting sequence or a signal peptide. The output of this first layer is then used in the second layer neural network as input to make the final prediction. Then the decision unit decides whether the cutoffs are obeyed. The output is then one of three classes cTP/mTP/SP/other and a reliability class value (RC) which is an indicator for the predictions certainty.

However, if a non-plant protein is entered the prediction for cTP is not applied for obvious reasons.

Predicted features

Predicts the localization to the following targets: chloroplast, mitochondrion, ER/golgi/secreted, and "other".

Required information for the prediction

The amino acid sequence of the protein and whether this protein is from a plant or non-plant organism.

Execution

Results and discussion

Task 3.4: Prediction of GO terms

GOPET

Details of the method

Author:

Year:

Reference:

Pfam

Details of the method

Author:

Year:

Reference:

ProtFun 2.2

Details of the method

Author:

Year:

Reference: