Sequence-based analyses of ARS A
Contents
1. Secondary Structure Prediction
Secondary structure prediction methods normally predict three different structural states:
- H (Helix): Helices are formed by short range interactions. The NH-group and the CO-group between two nearby amino acids form an H-bond and stabilize the structure. Three different H-bond patterns can be observed for the helical strucutere, which are interactions between amino acid i and i+3 ((3-10)-Helix), i+4 (alpha-Helix) and i+5 (Phi-Helix). As the stabilising interactions for this structural feature are near in sequence it is relatively easy to predict, even if only statistical properties are considered.
- E (beta-sheet): Beta Sheets are stabilised by long range interactions through the formation of H-bonds. As these interactions are between amino acids, that are not near in sequence, they are harder to predict than helices. Also here different patterns of H-bond formation can be observed. Parallel beta-sheets are formed by interactions between a residue i with the residues j-1 and j+1. Anti-parallel sheets are formed by interactions between residue i with j.
- C (coils): Coils are striuctural elements, which generally do not follow a recurrent pattern, i.e. they are untstructured. Normally these parts of proteins are rather flexible.
In the following two popular methods for secondary strucutre prediction are shortly intruduced and the applied to Arylsulfatase A. Predictions are then compared to the assignments of DSSP.
PSI-PRED
PSI-PRED was developed by David T. Jones in 1998. It requires a protein sequence in FASTA format. Then it performs a PSI-BLAST search and creates a sequence profile from the result. These profile capture evolutionary information about the set of proteins found and improves secondary structure prediction compared to the first or second generation methods, which only use statistcal properties of single amino acids or sliding windows.
The sequnce profile is then fed to a neural network with a feed-forward back-propagation architecture. The network consits of an input and output layer and a single hidden layer. The output of the first network then serves as input of a second network, which filters the prediction of the first network and yields the final prediction.
Together with the 3-state secondary structure prediction, PSI-PRED calculates confidence scores for the predicted structural elements. Thus, the user might identify false predictions within low-quality predicted regions.
The average Q3 score, reached by PSI-PRED is 80,3 %. <ref name="psipred">Jones, D. T.. "[Protein secondary structure prediction based on position-specific scoring matrices.]". J Mol Biol, 1999</ref>
Jpred
Jpred was developed by Cole, Barber and Barton in 1998. It also uses a neural network to predict secondary structure. The prediction relies on the Jnet algorithm, wich either takes a multiple sequence alignment or a single sequence as input. If a single sequence is passed to the program, Jpred also uses sequence profiles derived from a PSI-BLAST search.The output is presented as coloured HTML, plain text, PostScript, PDF and via the Jalview alignment editor. It also calculates confidence scores for all predicted residues to allow the user to identify possible false predictions. It reaches an average Q3 score up to 81,5 %. <ref name="jpred">Cole, C. and Barber, J. D. and Barton, G. J.. "[The Jpred 3 secondary structure prediction server.]". Nucleic Acid Res, 2008</ref>
DSSP
DSSP is a database of protein secondary structure assignments for all proteins in PDB. It was developed by Kabsch and Sander in 1983. It takes the 3D coordinates of a protein and assigns a hierarchical definition of secondary structure elements to the protein, based on different H-bond patterns. H-bonds are assigned below a specified cutoff, whereby the free energy is calculated using the Coulomb energy hydrogen bond model. <ref name="dssp">Kabsch W. and Sander C.. "[Dictionary of protein secondary structure: pattern recognition of hydrogen-bonded and geometrical features.]". Biopolymers, 1983</ref>
As stated above, DSSP assigns a hierachical definition of secondary structure and therefore the assignment contains more structural classes than the 3-state prediction (H=helix, E=sheet, C=coil) of PSI-PRED and Jpred. To be able to compare the predictions to the assignemnt of DSSP, its output must be converted to the three letter classification. We achieved this through writing a perl script. The following table depicts DSSP classes, their description and the "3-letter-class", we converted it to.
| DSSP class | Description ' | 3-letter class |
| H | Helix | H |
| G | 3-10 Helix | H |
| I | Phi-Helix | H |
| B | single bridge | E |
| E | beta sheet | E |
| T | turn | C |
| S | bend | C |
| \s | coil | C |
Results and Discussion
We predicted secondary structure of Arylsulfatase A with PSI-BLAST and Jpred3 using the Webserver user interface. Further on, we downloaded the DSSP secondary structure assignment and converted the hirachical definitions to the 3-state classification as described above. The predctions, together with the DSSP assignments are shown below.
Furthermore a textual representation of the predictions and assignments can be seen below. Missing residues in the DSSP output are marked by an "m".
mmmmmmmmmmmmmmmmmmCCCEEEEEEECCCCCCCCHHHCCCCCCCHHHHHHHHCCEEECCEECCCCCHHHHHHHHHHCCCHHHHCC (DSSP)
CCHHHHHHHHHHHCCCCCCCCCEEEEEEECCCCCCCCCCCCCCCCCHHHHHHHHCCCEECCCCCCCCCCHHHHHHHHHCCCCCCCCC (JPRED)
CCHHHHHHHHHHHHCCCCCCCCEEEEEECCCCCCCCCCCCCCCCCCCHHHHHHHCCCCCCCCCCCCCCCHHHHHHHHHCCCCCCCCC (PSI-PRED)
CCCCCCCCECCECCCCCCCHHHHHHCCCCEEEEEECCCCECCHHHCCCHHHHCCCEEEECCCCCCCCECCCCEEECCCEECCCCECC (DSSP)
CCCCCCCCCCCCCCCCCCHHHHHHHCCCCEEEEEECCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC (JPRED)
CCCCCCCCCCCCCCCCCCCHHHHHHHCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC (PSI-PRED)
CCCCCCEEECCEEEEECCCHHHHHHHHHHHHHHHHHHHHHCCCCEEEEEECCCCCCCCCCCCCCCCCCCCHHHHHHHHHHHHHHHHH (DSSP)
CCCCCCCCCCCCCCCCCCCCCCCCCHHHHHHHHHHHHCCCCCCCCCCCCCCCCCCCCCCCCCCCCCHHHHHHHHHHHHHHHHHHHHH (JPRED)
CCCCCCCCCCCCCCCCCCCCCCHHHHHHHHHHHHHHHHCCCCCCEEEEECCCCCCCCCCCCCCCCCCCCCCHHHHHHHHHHHHHHHH (PSI-PRED)
HHHHHHCCCHHHEEEEEEECCCCCHHHHHHCCCCCCCCCCCCCCCHHHHECCCEEECCCCCCCEEECCCEEHHHHHHHHHHHHCCCC (DSSP)
HHHHHHCCCCCCEEEEEEECCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCEEEEEECCCCCCCCCECCCCCCCCHHHHHHHHHCCCC (JPRED)
HHHHHHCCCCCCEEEEECCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCEEEEEEECCCCCCCCEECCCHHHHHHHHHHHHHHCCCC (PSI-PRED)
CCCCCCCCCCHHHHHCCCCCCCCEEEECCCCCCCCCCCCEEEECCEEEEEEECCCHHHCCCCCHHHCCCCCCEEEEEEEEEECCCCC (DSSP)
CCCCCCCCCCCCCCCCCCCCCCCEEEEECCCCCCCCCCEEEEECCCEEEECCCCCCCCCCCCCCCCCCCCCCCCCCCCCEECCCCCC (JPRED)
CCCCCCCCCCHHHHCCCCCCCCCEEEECCCCCCCCCCEEEEEECCCEEEEECCCCCCCCCCCCCCCCCCCCCCCCCCCEEEECCCCC (PSI-PRED)
CCCCCCCCCmmmCCHHHHHHHHHHHHHHHHHHHHCCCCCCCHHHCECHHHCCCCCCCCCCCCCCCCECmmmm (DSSP)
CCCCCCCCCCHHHHHHHHHHHHHHHHHHHHHCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC (JPRED)
CCCCCCCCCCCCCHHHHHHHHHHHHHHHHHHCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC (PSI-PRED)
Both methods show a good performance on the main part of the protein with an overall accurcy of 74 % for PSI-PRED and an accuracy of 71 % for Jpred3. All predicted sheets and helices highly overlap with assignments by DSSP. The errors of both predictions are mainly false negatives predictions of beta sheets. Further details of the prediction accuracy can be extracted from the table below:
| Program | #TP | #FP | accuracy |
| PSI-PRED | 374 | 133 | 0.74 |
| Jpred | 359 | 148 | 0.71 |
The accuracy (Q3) in these prediction is around 10 % lower than the average Q3 scores in the original publications of PSI-PRED and Jpred. Both methods predict the wrong secondary structure for the region from around position 110-200. Unfortunately the confidence scores generated by PSI-PRED and JPred are also rather high within this region, thus one is not able to identify this region as possibly false prediction from the prediction results alone.
DSSP assigns very short helices and beta sheets in this regions. Perhaps these are too short for a proper prediction.
2. Prediction of Disordered Regions
Disordered regions are regions in proteins that do not fold as expected and can be coil-like, globule-like, molten or something in between. In other words they can be defined definedregions of proteins that lack a fixed tertiary structure - i.e. are intrinsically unstructured. Often these disordered regions are important for binding and become ordered if the protei is associated with its cognate molecule.<ref>M.Rani, P.Romero, Z. Obradovic, A. Dunker. "Annotation of PDB with respect to "Disordered Regions" in Proteins" Download</ref>
Three different servers were challenged to predict disordered regions in ARSA, but no region was found that is consistently predicted disorderd by all three methods.
DISOPRED
DISOPRED was developed by David Jones et al. in 2004. DISOPRED2 uses Support Vector Machines for Disoredered region prediction. It was trained on a set of around 750 non-redundant sequences with high resolution X-ray structures. As disordered regions cannot easily be determined experimental methods, disorder is simply assigned to missing regions in the resolved structure. For the training and classification of the SVM, sequence profiles are used. <ref name="disopred">Ward, J. J. and McGuffin, L. J. and Bryson, K. and Buxton, B. F. and Jones, D. T. "[The DISOPRED server for the prediction of protein disorder.]". Bioinformatics, 2004</ref>
We used the DISOPRED server for the prediction:
DISOPRED predictions for a false positive rate threshold of: 2%
conf: 930000000000012210000000000000000000000000000000000000000000
pred: *...........................................................
AA: MGAPRSLLLALAAGLAVARPPNIVLIFADDLGYGDLGCYGHPSSTTPNLDQLAAGGLRFT
10 20 30 40 50 60
conf: 000000000000000000000000000000000000000000000000000000000000
pred: ............................................................
AA: DFYVPVSLCTPSRAALLTGRLPVRMGMYPGVLVPSSRGGLPLEEVTVAEVLAARGYLTGM
70 80 90 100 110 120
conf: 000000000000000000000000000000000000000000000000000000000000
pred: ............................................................
AA: AGKWHLGVGPEGAFLPPHQGFHRFLGIPYSHDQGPCQNLTCFPPATPCDGGCDQGLVPIP
130 140 150 160 170 180
conf: 000000000000000000000000000000000000000000000000000000000000
pred: ............................................................
AA: LLANLSVEAQPPWLPGLEARYMAFAHDLMADAQRQDRPFFLYYASHHTHYPQFSGQSFAE
190 200 210 220 230 240
conf: 000000000000000000000000000000000000000000000000000000000000
pred: ............................................................
AA: RSGRGPFGDSLMELDAAVGTLMTAIGDLGLLEETLVIFTADNGPETMRMSRGGCSGLLRC
250 260 270 280 290 300
conf: 000000000000000000000000000000000000000000000000000000000000
pred: ............................................................
AA: GKGTTYEGGVREPALAFWPGHIAPGVTHELASSLDLLPTLAALAGAPLPNVTLDGFDLSP
310 320 330 340 350 360
conf: 000000000000000000000000000000000000000000000000000000000000
pred: ............................................................
AA: LLLGTGKSPRQSLFFYPSYPDEVRGVFAVRTGKYKAHFFTQGSAHSDTTADPACHASSSL
370 380 390 400 410 420
conf: 000000000000000000000000000000000000000000000000000000000000
pred: ............................................................
AA: TAHEPPLLYDLSKDPGENYNLLGGVAGATPEVLQALKQLQLLKAQLDAAVTFGPSQVARG
430 440 450 460 470 480
conf: 000000000000000002571699999
pred: ......................*****
AA: EDPALQICCHPGCTPRPACCHCPDPHA
490 500
Asterisks (*) represent disorder predictions and dots (.)
prediction of order. The confidence estimates give a rough
indication of the probability that each residue is disordered.
As we already know the structure of ARSA, we do not expect long disordered regions. Possibly, only short loops might be identified as unstructured elements by DISOPRED. As you can see, only the first residue and the five last residues are predicted to be in a disordered region with very high confidence.
This prediction makes sense, if we consider the DSSP assignment for this region. DSSP assigns coils for the last residues of the protein, which are secondary structure alements which do not follow specific structural patterns by definition. The rest of the coils within the protein is probably not predicted to be unstructured because they are located within the core and thus do not allow structural flexibility.
POODLE
POODLE predicts many disordered residues. Depending on the treshold one can identify 6 or more disordered regions. Like DISOPRED it predicts Disorder at the end of the protein. Please see the previous section for a discussion.
The disorder prediction for the beginning also makes sense, as DSSP assigns a loop region there. The disorder predictions at residues 150-200 and 400-450 are clearly false as the overlap with helices and beta-sheets. Thus these regions are structured.
IUPred
IUPred was developed by Simon et al. in 2005. <ref name="iupred">Zsuzsanna Dosztanyi, Veronika Csizmak, Peter Tompa and Istvan Simon . "[The Pairwise Energy Content Estimated from Amino Acid Composition Discriminates between Folded and Intrinsically Unstructured Proteins .]". J Mol Biol, 2005</ref>
It estimates the total pairwise interaction energy from the amino acid composition of the protein. This energy space can clearly discriminate between disordered and structured regions.
The three different options of prediction were tried and are illustrated below. In general, IUPred did not predict any disordered region with a "Disorder tendency" above 0.6 except one very short region around residue 415 with the "long disorder"-option.
long disorder
The main profile of our server is to predict context-independent global disorder that encompasses at least 30 consecutive residues of predicted disorder. For this application the sequential neighbourhood of 100 residues is considered. <ref name="IUPred"> http://iupred.enzim.hu/Help.html</ref>
short disorder
It uses a parameter set suited for predicting short, probably context-dependent, disordered regions, such as missing residues in the X-ray structure of an otherwise globular protein. For this application the sequential neighbourhood of 25 residues is considered. As chain termini of globular proteins are often disordered in X-ray structures, this is taken into account by an end-adjustment parameter which favors disorder prediction at the ends. <ref name="IUPred"> http://iupred.enzim.hu/Help.html</ref>
structured domains
The dependable identification of ordered regions is a crucial step in target selection for structural studies and structural genomics projects. Finding putative structured domains suitable for stucture determination is another potential application of this server. In this case the algorithm takes the energy profile and finds continuous regions confidently predicted ordered. Neighbouring regions close to each other are merged, while regions shorter than the minimal domain size of at least 30 residues are ignored. When this prediction type is selected, the region(s) predicted to correspond to structured/globular domains are returned. <ref name="IUPred"> http://iupred.enzim.hu/Help.html</ref>
Summary
No mode of IUPred predicts any disordered region. This makes perfectly sense as ARSA does not contain disordered region, if we disregard loops.
Meta-Disorder
We used the PredictProtein Webserver for the prediction:
The result is shown below ("D"=disordered, "-"=not disordered)
----------------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------------
-------
Also Meta-Disorder did not predict any disordered regions for this ARSA.
Discussion
The only server that predicted significant disordered regions was POODLE. However the predictions by POODLE for the highly overlap with secondary structure elements of the protein, thus they are wrong. The other methods agreed in the result that no disordered regions can be found except for the ends of the sequence, which makes sense as the end of the protein consists of an unstructured loop.
3. Prediction of transmembrane alpha-helices and signal peptides
The prediction of membrane proteins and their topology is very important, because the experimental determination of these protein is quite challenging. It is very dificult to determine the structure, because the influence of membrane mimetic environments might lead to non-native structures and thus lead to a wrongf structural model of the protein. <ref>Cross, Timothy, Mukesh Sharma, Myunggi Yi, Huan-Xiang Zhou (2010). "Influence of Solubilizing Environments on Membrane Protein Structures"</ref>
Additional Proteins
The following proteins are additionally used for the prediction of transmembrand alpha-helices and signal peptides and for the prediction of GO Terms:
BACR
BACR_HALSA is a bacterial membrane protein. It is involved in hydrogen ion transport and sensory transduction transport. The topology of cellular and transmembrane domains of the protein is shown below:
| type | Position ' | Description |
| Topological domain | 14 – 23 | Extracellular |
| Transmembrane | 24 – 42 | Helical; Name=Helix A |
| Topological domain | 43 – 56 | Cytoplasmic |
| Transmembrane | 57 – 75 | Helical; Name=Helix B |
| Topological domain | 76 – 91 | Extracellular |
| Transmembrane | 92 – 109 | Helical; Name=Helix C |
| Topological domain | 110 – 120 | Cytoplasmic |
| Transmembrane | 121 – 140 | Helical; Name=Helix D |
| Topological domain | 141 – 147 | Extracellular |
| Transmembrane | 148 – 167 | Helical; Name=Helix E |
| Topological domain | 168 – 185 | Cytoplasmic |
| Transmembrane | 186 – 204 | Helical; Name=Helix F |
| Topological domain | 205 – 216 | Extracellular |
| Transmembrane | 217 – 236 | Helical; Name=Helix G |
| Topological domain | 237 – 262 | Cytoplasmic |
RET 4
- RET4_HUMAN is a human retinal-binding protein. It delivers retinol from the liver stores to the peripheral tissues. Defects can cause night vision problems.
INSL 5
- INSL5_HUMAN is a human insulin-like peptide. It consists of two chains and may have a role in gut contractility or in thymic development and regulation.
LAMP 1
- LAMP1_HUMAN is a human membrane glycoprotein. It presents cabohydrate ligands to selectins. It is located in the lysosomal membrane. Its topology is shown below:
| type | Position ' | Description |
| Topological Domain | 29 - 382 | Lumenal |
| Transmembrane | 383 - 405 | Helical |
| Topological Domain | 406 - 417 | Cytoplasmic |
| Region | 29 - 194 | First lumenal domain |
| Region | 195 - 227 | Hinge |
| Region | 228 - 382 | Second lumenal domain |
A 4
- A4_HUMAN is a human cell surface receptor involved in neurite growth, neuronal adhesion and axonogenesis. It is involved in Alzheimer disease and Amyloidosis.
| type | Position ' | Description |
| Topological domain | 18 - 699 | Extracellular |
| Transmembrane | 700 - 723 | Helical |
| Topological domain | 724 - 770 | Cytoplasmic |
| Domain | 291 - 341 | BPTI / Kunitz inhibitor |
| Region | 96 - 110 | Heparin-binding |
| Region | 181 - 188 | Zinc-binding |
| Region | 391 - 423 | Heparin-binding |
| Region | 491 - 522 | Heparin-binding |
| Region | 523 - 540 | Collagen-binding |
| Region | 732 - 751 | Interaction with G(o)-alpha |
| Motif | 724 - 734 | Basolateral sorting signal |
| Motif | 759 - 762 | NPXY motif; contains endocytosis signal |
| Compositional bias | 230 - 260 | Asp/Glu-rich (acidic) |
| Compositional bias | 274 - 280 | Poly-Thr |
SignalP
SignalP uses a neural network and a HMM to calculate three different scores <ref name="signalp">Bendtsen, J. D. and Nielsen, H. and von Heijne, G. and Brunak, S.. "[Improved prediction of signal peptides: SignalP 3.0.]". J Mol Biol, 2004</ref>:
- S-score (=signal peptide score): High values indicate the presence of a signal peptides in the sequence.
- C-Score (=raw cleavage site score): This score is used to recognize the cleavage site.
- Y-score (= combined cleavage site score): This score optimizes the prediction of the cleavage site by considering the C-score and the S-score simultaneously. A cleavage site is predicted, if the C-score is high and the S-score is low.Optimierung des cleavage site scores.
We executed SignalP with the following commands:
sudo /apps/signalp-3.0/signalp -t gram- ../BACR.fasta > BACR.signalp
sudo /apps/signalp-3.0/signalp -t euk ../ARSA.fasta > ARSA.signalp
sudo /apps/signalp-3.0/signalp -t euk ../A4.fasta > A4.signalp
sudo /apps/signalp-3.0/signalp -t euk ../LAMP1.fasta > LAMP1.signalp
sudo /apps/signalp-3.0/signalp -t euk ../INSL5.fasta > INSL5.signalp
sudo /apps/signalp-3.0/signalp -t euk ../RET4.fasta > RET4.signalp
The graphical output of the method is shown below:
| ARS A | A4 | RET4 | INSL5 | LAMP1 | BACR |
|
|
|
|
|
|
Discussion
The cleavage sites and signal peptides of all proteins are correctly predicted, compard to the UniprotKB annotation. In general the output of the neural network gives a more distinct prediction of the different regions. The bacterial membrane protein BACR does not contain a signal peptide, regarding the annotation of UniprotKB and SignalP does not predict one, but the S-score is very high between position 20-40, which is a transmembrane helix. This is due to the similar properties of signal peptides and transmembrane helices, which both exhibit a bias towards hydrophobic amino acids. But lacking the characteristics of a cleavage site, SignalP does not predict a Signalpeptide here. This shows that the program is able to properly distinguish between transmebrane helices and signal peptides.
TMHMM
TMHMM predicts transmembrane helices (TMH) using a Hidden Markov Model (HMM). The protein described by TMH model essentially consists of seven different states. Globular domains can occur on the cytoplasmic and the non-cytoplasmic side. On the cytoplsmic side, globular domains are linked to loops, ehich are agin linked to cytoplasimc caps. These caps are followed by the helex core and there is again a cap on the non-cytoplasmic side. These caps are linked to globular domains by either short or long non-cytoplasmic loops.
TMHMM outputs the most likely structure of the protein, ragarding to the above model. It also includes the orientation (cytoplasmic or non-cytoplasmic side) of the N-terminal signal sequence. The ouput consists of a plot - graphically showing the different states along the protein - and some additional statistics <ref> http://www.cbs.dtu.dk/services/TMHMM-2.0/TMHMM2.0.guide.html#output </ref>:
- The number of predicted transmembrane helices.
- The expected number of amino acids in transmembrane helices. If this number is larger than 18 it is very likely to be a transmembrane protein (OR have a signal peptide).
- The expected number of amino acids in transmembrane helices in the first 60 amino acids of the protein. If this number more than a few, you should be warned that a predicted transmembrane helix in the N-term could be a signal peptide.
- The total probability that the N-term is on the cytoplasmic side of the membrane.
| Protein | #predicted TMHs | #expected AAs in TMHs | #expected AAs in TMHs in first 60 positions | orientation (N-term at non-cyto. side) | Graphical output |
| ARS A | 0 | 2.65106 | 2.63079 | 0.12149 | 
|
| A4_HUMAN | 1 | 22.72525 | 0.0027 | 0.00015 | 
|
| INSL5_HUMAN | 0 | 0.50415 | 0.50415 | 0.03772 | 
|
| LAMP1_HUMAN | 2 | 44.89582 | 22.24286 | 0.99287 | 
|
| RET4_HUMAN | 0 | 0.01196 | 0.01179 | 0.01909 | 
|
| BACR | 6 | 140.4032 | 26.1196 | 0.01887 | 
|
Discussion
- ARS A: All amino acids are predicted to be "outside" the membrane, which is consistent with the UniprotKB annotation, as ARS A is not a membrane protein. The graphical output of TMHMM shows, that the probaility for a transmembrane helix is elevated at the start of the protein, which is due to the hydrophobicity of the signal peptide.
- A4_HUMAN: This protein contains exactly one transmembrane helix which is located from postion 700-723. TMHMM predicts the transmembrane helix at 701-723, which is quite stifying. The predited topology is given below:
| Description | Position ' |
| outside | 1-700 |
| TMhelix | 701-723 |
| inside | 724-770 |
- INSL5_HUMAN: This protein does not contain any transmembrane helices and none are predicted by TMHMM.
- LAMP1_HUMAN: TMHMM predicts a Possible N-terminal signal sequence and two potential transmembrane helices. LAMP1 indeed contains a N-terminal signal peptide, but the prediction of the first transmembrane helix is false. This false positive prediction overlaps to 50% with the signal peptide, which is located from position 1-22. However, the second predicted TM-helix highly overlaps with the annotated transmebrane helix in UniprotKB.
| Description | Position ' |
| inside | 1-10 |
| TMhelix | 11-33 |
| outside | 34-383 |
| TMhelix | 384-406 |
| inside | 407-417 |
- RET4_HUMAN: No TM-helices are predicted, which coincides with the annotation.
- BACR: A Possible N-terminal signal sequence is predicted, which is false. BACR contains 7 TM-helices. TMHMM only predicts 6 transmembrane helices, which highly overlap with the annotation. The prediction misses the last TM-helix in the protein. Despite the false prediction, the graphical output shows, that the probybility is quite high in this region.
| Description | Position ' |
| outside | 1-22 |
| TMhelix | 23-42 |
| inside | 43-54 |
| TMhelix | 55-77 |
| outside | 78-91 |
| TMhelix | 92-114 |
| inside | 115-120 |
| TMhelix | 121-143 |
| outside | 144-147 |
| TMhelix | 148-170 |
| inside | 171-189 |
| TMhelix | 190-212 |
| outside | 213-262 |
Phobius and Polyphobius
Phobius combines SignalP-HMM and TMHMM for the prediction of transmembrane proteins and their topology. The Hidden Markov Models of both progams are simply associated via the last state of SignalP-HMM and the non-cytoplasmic loop state of TMHMM. This is justified, because most signal peptides are located at the non-cytoplasmic side of the membrane, but it also limits the detection of proteins with the opposite location. <ref name="phobius">Kall, L. and Krogh, A. and Sonnhammer, E. L.. "[A combined transmembrane topology and signal peptide prediction method.]". J Mol Biol, 2004</ref>
Polyphobius extends the approach of Phobius by incorporating information from homologs using global alignments. <ref name="polyphobius">Kall, L. and Krogh, A. and Sonnhammer, E. L.. "[An HMM posterior decoder for sequence feature prediction that includes homology information.]". Bioinformatics, 2005</ref>
| Protein | Phobius - graphical | Phobius - textual | Polyphobius - graphical | Polyphobius - textual |
| A4 | 
|
FT SIGNAL 1 17 FT REGION 1 1 N-REGION. |
|
FT SIGNAL 1 17 FT REGION 1 3 N-REGION. |
| ARS A | 
|
FT SIGNAL 1 28 FT REGION 1 6 N-REGION. |
|
FT SIGNAL 1 16 FT REGION 1 4 N-REGION. |
| BACR | 
|
FT TOPO_DOM 1 22 NON CYTOPLASMIC. FT TRANSMEM 23 42 |
|
FT TOPO_DOM 1 21 NON CYTOPLASMIC. FT TRANSMEM 22 43 |
| INSL5 | 
|
FT SIGNAL 1 22 FT REGION 1 5 N-REGION. |
|
FT SIGNAL 1 22 FT REGION 1 4 N-REGION. |
| LAMP1 | 
|
FT SIGNAL 1 28 FT REGION 1 10 N-REGION. |
|
FT SIGNAL 1 28 FT REGION 1 9 N-REGION. |
| RET4 | 
|
FT SIGNAL 1 18 FT REGION 1 2 N-REGION. |
|
FT SIGNAL 1 18 FT REGION 1 3 N-REGION. |
Discussion
- ARS A: The prediction of the signal peptide of Phobius is too long (1-28), whereas the prediction of Polyphobius is slighlty too short (1-16). Regarding to the annotaion, ARS A contains a signal peptide from position 1-18. Both methods don't predict a transmembrane helix.
- A4_HUMAN: Both methods correctly predict the location of the signal peptide. The prediction of the transmembrane helix only misses the first amino acid. This prediction is quite good.
- INSL5_HUMAN: Phobius and Polyphobius correctly predict the signal peptide from position 1-22 and no TM-helix.
- LAMP1_HUMAN: Both methods correctly predict the location of the signal peptide. The prediction of the transmembrane helix contains an additional amino acid at the start.
- RET4_HUMAN: Phobius and Polyphobius correctly predict the signal peptide from position 1-18 and no TM-helix.
- BACR: Both methods predict 7 transmembrane helices, which almost perfectly overlap with the annotation.
Phobius and Polyphobius yield very similar results. The only improvement for Polyphobius - which can be seen from our analysed proteins - is in the prediction of the location of the signal peptide for ARS A.
OCTOPUS and SPOCTOPUS
OCPTOPUS uses a combination of a Hidden Markov Model and neural network to predict the topology of a transmembrane protein. It uses BALST to create a sequence profile, whihc is then used by the neural network to predict the preference of the amino acids to be located within a transmembrane (M), interface (I), close loop (L) globular loop (G), inside (i) or outside (o). These scores are then passed to the HMM, which predicts the final states. <ref name="octopus">Viklund, H. and Elofsson, A.. "[OCTOPUS: improving topology prediction by two-track ANN-based preference scores and an extended topological grammar.]" Bioinformatics, 2008</ref>
SPOCTOPUS extends the OCTOPUS algorithm with a preprocessing step. OCTOPUS does not predict signal peptides. The N-terminal targeting sequences mainly consist of hydrophobic residues and thus thier properties strongly resemble the transmembrane helices. Not considering the signal peptides in the prediction often leads to a false prediction of a transmembrane helix at the N-terminal domain. Therefore SPOCTOPUS extends the OCTOPUS algorithm with the prediction of signal peptide preference scores within the first 70 amino acids of the protein. The exact location of a potential signal peptide are then predicted by a HMM in OCTOPUS. <ref name="spoctopus">Viklund, H. and Bernsel, A. and Skwark, M. and Elofsson, A.. "[SPOCTOPUS: a combined predictor of signal peptides and membrane protein topology.]" Bioinformatics, 2008</ref>
| Protein | OCTOPUS | SPOCTOPUS |
| ARS A | 
|
|
| A4 | 
|
|
| RET4 | 
|
|
| INSL5 | 
|
|
| LAMP1 | 
|
|
| BACR | 
|
|
Discussion
- ARS A: OCTOPUS does not predict any TM-regions or the signal peptide. SPOCTOPUS correctly predicts the signal peptide.
- A4_HUMAN: Both methods correctly predict the location of the TM-helix. SPOCTOPUS also correctly predicts the signal peptide. OCTOPUS predicts a REENTRANT/DIP region at the end of the signal peptide.
- RET4_HUMAN: OCTOPUS confounds the signal peptide with a TM-helix. Contrary, SPOCTOPUS correctly predicts a signal peptide.
- INSL5_HUMAN: OCTOPUS confounds the signal peptide with a TM-helix. Contrary, SPOCTOPUS correctly predicts a signal peptide.
- LAMP1_HUMAN: Both methods correctly predict the location of TM-helix. Again, OCTOPUS confounds the signal peptide with a TM-helix, but SPOCTOPUS yields a correct prediction.
- BACR: Both methods predict 7 transmembrane helices, which almost perfectly overlap with the annotation.
The difference between OCTOPUS and SPOCTOPUS can be clearly seen in the predictions. As mentioned above, OCTOPUS does not include the prediction of signal peptides and thus confounds signal peptides with TM-helices.
TargetP
TargetP is used to predict the cellular localization of a protein. It combines the two methods ChloroP and SignalP. The following targeting sequences can be identified:
- chloroplast transit peptide (cTP)
- mitochondrial targeting peptide (mTP)
- secretory pathway signal peptide (SP)
TargetP uses a neural network to calculate and outputs scores for each of the above subcellular targets. TargetP finally predicts the location with the highest score. In our case all proteins are predicted to be targeted to the secretory pathway (S). Results are shown below. Note, that cTP is not included in our predictions, as we only considered eukaryotic and bacterial proteins. Also note, that TargetP is trained on eukaryotic proteins and hence the prediction for the protein "BACR", which is bacterial does not make sense, because there are completely different pathways of localization and secretion in eukayotes and bacteria (e.g. bacteria do not have an endoplasmatic reticulum, Golgi-Apparatus or Lysosome). Nevertheless, we included it in our analysis to see if TargetP predicts finds any localization sequence in it or predicts "-" (= no localization signal found). <ref name="targetp">Emanuelsson, O. and Nielsen, H. and Brunak, S. and von Heijne, G.. "[Predicting subcellular localization of proteins based on their N-terminal amino acid sequence.]". J Mol Biol, 2000</ref>
Our prediction results are dpicted in the following table:
| Protein | mTP | SP | other | prediction |
| ARS A | 0.070 | 0.926 | 0.054 | S |
| A4_HUMAN | 0.035 | 0.937 | 0.084 | S |
| INSL5_HUMAN | 0.074 | 0.899 | 0.037 | S |
| LAMP1_HUMAN | 0.043 | 0.953 | 0.017 | S |
| RET4_HUMAN | 0.242 | 0.928 | 0.020 | S |
| BACR (bacterial) | 0.019 | 0.897 | 0.562 | S |
Discussion
All proteins are assigned to the secretory pathway.
- Arylsulfatase A is a lysosomal enzyme. Therefore, the prediction is correct, as lysosomal proteins are guided there by the secretory pathway, via the endoplasmatic reticulum and the Golgi apparatus.
- A4 is a membrane protein. Thus the prediction is correct.
- INSL5 is also a secreted (regarding UniprotKB) and the prediction is correct.
- LAMP1 is targeted to the cellular membrane. Thus the prediction is correct.
- RET4 delivers retinol from the liver stores to the peripheral tissues. It is a secretory protein, thus the prediciton is correct.
- As described above, BACR is a bacterial protein. TargetP assigns, that this protein is also guided to the secretory pathway, which makes no sense as the bacterial protein secretion is different from eukaryotic secretion. Nevertheless, the prediction is much less obvious in this case, compared to the others. The "other" class - meaning that no targeting sequence is found in the protein gets a considerable high score in this prediction, hence the assignment to S is more questionable here.
4. Prediction of GO Terms
The aim of the gene ontology is to standardize the representation of genes and gene products. The gene ontology is divided into three major parts: cellular component, molecular function and biological process. For us, the molecular function is interesting as it describes activities of a protein. Each GO-Term stands for a specific function, so predicting GO-Terms means prediction of protein-function. <ref>http://www.geneontology.org</ref>
GOPET
GOPET uses homology searches and a SVM to predict GO-Terms.
GO-Terms for 6 different proteins were predicted. The results are shown below. Bold entries are GO-Terms which are really connected to the protein. <ref>http://www.ebi.ac.uk/QuickGO/</ref>
A4
| GOid | Confidence | GO term |
| GO:0004866 | 87% | endopeptidase inhibitor activity |
| GO:0004867 | 86% | serine-type endopeptidase inhibitor activity |
| GO:0030568 | 83% | plasmin inhibitor activity |
| GO:0030304 | 83% | trypsin inhibitor activity |
| GO:0030414 | 82% | peptidase inhibitor activity |
| GO:0005488 | 79% | binding |
| GO:0005515 | 74% | protein binding |
| GO:0046872 | 73% | metal ion binding |
| GO:0003677 | 71% | DNA binding |
| GO:0008201 | 70% | heparin binding |
| GO:0008270 | 69% | zinc ion binding |
| GO:0005507 | 69% | copper ion binding |
| GO:0005506 | 67% | iron ion binding |
ARS A
| GOid | Confidence | GO term |
| GO:0003824 | 97% | catalytic activity |
| GO:0016787 | 96% | hydrolase activity |
| GO:0008484 | 95% | sulfuric ester hydrolase activity |
| GO:0004065 | 92% | arylsulfatase activity |
| GO:0004098 | 89% | cerebroside-sulfatase activity |
| GO:0003943 | 83% | N-acetylgalactosamine-4-sulfatase activity |
| GO:0004773 | 82% | steryl-sulfatase activity |
| GO:0004423 | 82% | iduronate-2-sulfatase activity |
| GO:0008449 | 82% | N-acetylglucosamine-6-sulfatase activity |
| GO:0047753 | 82% | choline-sulfatase activity |
| GO:0018741 | 81% | alkyl sulfatase activity |
| GO:0046872 | 63% | metal ion binding |
| GO:0016250 | 61% | N-sulfoglucosamine sulfohydrolase activity |
BACR_HALSA
| GOid | Confidence | GO term |
| GO:0005216 | 77% | ion channel activity |
| GO:0008020 | 75% | G-protein coupled photoreceptor activity |
| GO:0015078 | 60% | hydrogen ion transmembrane transporter activity |
INSL 5
| GOid | Confidence | GO term |
| GO:0005179 | 80% | hormone activity |
LAMP 1
| GOid | Confidence | GO term |
| GO:0004812 | 60% | aminoacyl-tRNA ligase activity |
| GO:0005524 | 60% | ATP binding |
RET 4
| GOid | Confidence | GO term |
| GO:0005488 | 90% | binding |
| GO:0005501 | 81% | retinoid binding |
| GO:0008289 | 80% | lipid binding |
| GO:0019841 | 78% | retinol binding |
| GO:0005215 | 78% | transporter activity |
| GO:0016918 | 78% | retinal binding |
| GO:0005319 | 69% | lipid transporter activity |
| GO:0008035 | 60% | high-density lipoprotein particle binding |
Summary
In most cases, the Predictions of GOPET give a good hint on the function of the protein. For
- A4 8 out of 14
- ARSA 6 out of 13
- BACR 1 out of 3
- INSL5 1 out of 1
- LAMP1 0 out of 1
- RET4 5 out of 8
predictions were truely related to the function of the proteins. Due to the use of sequence searches and thus homology information the method will perform better for well characterized protein families. Predictions for new, previously unknown families will probably yield more false predictions.
Pfam
Pfam is a large database, which stores protein families, which are represented by multiple sequence alignments and hidden Markov models. We performed sequence searches of the proteins of interest and extracted the GO Terms associated with the resulting families from the database.
GO-Terms for 6 different proteins were predicted. The results are shown below. The results were mapped to GO-terms by the list pfam2go <ref>http://www.geneontology.org/external2go/pfam2go</ref> Bold entries are GO-Terms which are really connected to the protein. <ref>http://www.ebi.ac.uk/QuickGO/</ref>
A4
| PFAM-Family | PFAM-Description | GOid | GO term |
| APP_N | Amyloid-A4 N-terminal heparin-binding | ||
| APP_Cu_bd | Copper-binding of amyloid precursor, CuBD | ||
| Kunitz_BPTI | Kunitz / Bovine pancreatic trypsin inhibitor domain | GO:0004867 | serine-type endopeptidase inhibitor activity |
| APP_E2 | E2 domain of amyloid precursor protein | ||
| Beta-APP | Beta-amyloid peptide (Beta-APP) | ||
| APP_amyloid | beta-amyloid precursor protein C-terminus |
ARS A
| PFAM-Family | PFAM-Description | GOid | GO term |
| Sulfatase | Sulfatase | GO:0008484 | sulfuric ester hydrolase activity |
BACR HALSA
| PFAM-Family | PFAM-Description | GOid | GO term |
| Bac_rhodopsin | Bacteriorhodopsin-like protein | GO:0005216, GO:0006811, GO:0016020 | ion channel activity, ion transport, membrane |
INSL 5
| PFAM-Family | PFAM-Description | GOid | GO term |
| Insulin | Insulin / IGF / Relaxin family | GO:0005179, GO:0005576 | hormone activity, extracellular region |
LAMP 1
| PFAM-Family | PFAM-Description | GOid | GO term |
| Lamp | Lysosome-associated membrane glycoprotein (Lamp) | GO:0016020 | membrane |
RET 4
| PFAM-Family | PFAM-Description | GOid | GO term |
| Lipocalin | Lipocalin / cytosolic fatty-acid binding protein family | GO:0005488 | binding |
Summary
The predictions using Pfam yield much less results, compared to GOPET. For
- A4 1 out of 6
- ARSA 1 out of 1
- BACR 1 out of 1
- INSL5 1 out of 1
- LAMP1 1 out of 1
- RET4 1 out of 1
predictions were truely related to the function of the proteins. As in this method, also homology information is needed, the predictions for new, previously unknown families will probably yield more false predictions.
ProtFun 2.2
ProtFun queries other feature prediction servers and integrates all the results in its final prediction.
A4
############## ProtFun 2.2 predictions ############## >sp_P05067_A # Functional category Prob Odds Amino_acid_biosynthesis 0.020 0.921 Biosynthesis_of_cofactors 0.261 3.623 Cell_envelope => 0.804 13.186 Cellular_processes 0.053 0.730 Central_intermediary_metabolism 0.184 2.920 Energy_metabolism 0.023 0.259 Fatty_acid_metabolism 0.016 1.265 Purines_and_pyrimidines 0.417 1.716 Regulatory_functions 0.013 0.084 Replication_and_transcription 0.029 0.109 Translation 0.027 0.613 Transport_and_binding 0.827 2.016 # Enzyme/nonenzyme Prob Odds Enzyme => 0.392 1.368 Nonenzyme 0.608 0.852 # Enzyme class Prob Odds Oxidoreductase (EC 1.-.-.-) 0.024 0.114 Transferase (EC 2.-.-.-) 0.208 0.603 Hydrolase (EC 3.-.-.-) 0.190 0.600 Lyase (EC 4.-.-.-) 0.020 0.430 Isomerase (EC 5.-.-.-) 0.010 0.324 Ligase (EC 6.-.-.-) 0.048 0.946 # Gene Ontology category Prob Odds Signal_transducer 0.126 0.586 Receptor 0.036 0.211 Hormone 0.001 0.206 Structural_protein => 0.034 1.205 Transporter 0.024 0.222 Ion_channel 0.009 0.162 Voltage-gated_ion_channel 0.002 0.108 Cation_channel 0.010 0.215 Transcription 0.043 0.335 Transcription_regulation 0.018 0.143 Stress_response 0.076 0.862 Immune_response 0.016 0.183 Growth_factor 0.005 0.372 Metal_ion_transport 0.009 0.020 //
It was not possible to map this ProtFun-result to a Gene Ontology category.
ARS A
############## ProtFun 2.2 predictions ############## >sp_P15289_A # Functional category Prob Odds Amino_acid_biosynthesis 0.015 0.669 Biosynthesis_of_cofactors 0.048 0.668 Cell_envelope => 0.804 13.186 Cellular_processes 0.027 0.373 Central_intermediary_metabolism 0.404 6.416 Energy_metabolism 0.050 0.555 Fatty_acid_metabolism 0.028 2.138 Purines_and_pyrimidines 0.404 1.662 Regulatory_functions 0.013 0.081 Replication_and_transcription 0.021 0.080 Translation 0.032 0.717 Transport_and_binding 0.821 2.002 # Enzyme/nonenzyme Prob Odds Enzyme => 0.540 1.886 Nonenzyme 0.460 0.644 # Enzyme class Prob Odds Oxidoreductase (EC 1.-.-.-) 0.063 0.304 Transferase (EC 2.-.-.-) 0.062 0.180 Hydrolase (EC 3.-.-.-) 0.313 0.987 Lyase (EC 4.-.-.-) 0.038 0.803 Isomerase (EC 5.-.-.-) 0.010 0.321 Ligase (EC 6.-.-.-) 0.017 0.326 # Gene Ontology category Prob Odds Signal_transducer 0.206 0.965 Receptor 0.111 0.652 Hormone 0.002 0.323 Structural_protein 0.005 0.177 Transporter 0.025 0.229 Ion_channel 0.009 0.154 Voltage-gated_ion_channel 0.003 0.139 Cation_channel 0.010 0.215 Transcription 0.037 0.287 Transcription_regulation 0.018 0.142 Stress_response => 0.102 1.158 Immune_response 0.022 0.259 Growth_factor 0.005 0.391 Metal_ion_transport 0.009 0.020 //
It was not possible to map this ProtFun-result to a Gene Ontology category.
BACR HALSA
############## ProtFun 2.2 predictions ############## >sp_P02945_B # Functional category Prob Odds Amino_acid_biosynthesis 0.033 1.495 Biosynthesis_of_cofactors 0.186 2.589 Cell_envelope 0.029 0.483 Cellular_processes 0.051 0.694 Central_intermediary_metabolism 0.045 0.711 Energy_metabolism 0.138 1.537 Fatty_acid_metabolism 0.016 1.265 Purines_and_pyrimidines 0.302 1.244 Regulatory_functions 0.013 0.080 Replication_and_transcription 0.019 0.073 Translation 0.059 1.339 Transport_and_binding => 0.791 1.929 # Enzyme/nonenzyme Prob Odds Enzyme 0.199 0.696 Nonenzyme => 0.801 1.122 # Enzyme class Prob Odds Oxidoreductase (EC 1.-.-.-) 0.114 0.549 Transferase (EC 2.-.-.-) 0.031 0.091 Hydrolase (EC 3.-.-.-) 0.057 0.180 Lyase (EC 4.-.-.-) 0.020 0.430 Isomerase (EC 5.-.-.-) 0.010 0.321 Ligase (EC 6.-.-.-) 0.017 0.326 # Gene Ontology category Prob Odds Signal_transducer 0.258 1.205 Receptor 0.355 2.087 Hormone 0.001 0.206 Structural_protein 0.006 0.200 Transporter => 0.440 4.036 Ion_channel 0.010 0.169 Voltage-gated_ion_channel 0.004 0.172 Cation_channel 0.078 1.689 Transcription 0.026 0.205 Transcription_regulation 0.028 0.226 Stress_response 0.012 0.139 Immune_response 0.011 0.128 Growth_factor 0.010 0.727 Metal_ion_transport 0.049 0.106 //
"Transporter" corresponds to GO:0005215 which is not annotated in http://www.ebi.ac.uk/QuickGO/GProtein?ac=P02945 but if you look closer you can see that BACR_HAlSA is a ion transporter, so the classification is true.
INSL 5
############## ProtFun 2.2 predictions ############## >sp_Q9Y5Q6_I # Functional category Prob Odds Amino_acid_biosynthesis 0.011 0.484 Biosynthesis_of_cofactors 0.040 0.558 Cell_envelope => 0.756 12.393 Cellular_processes 0.033 0.448 Central_intermediary_metabolism 0.048 0.755 Energy_metabolism 0.036 0.397 Fatty_acid_metabolism 0.016 1.265 Purines_and_pyrimidines 0.144 0.592 Regulatory_functions 0.014 0.087 Replication_and_transcription 0.020 0.075 Translation 0.032 0.735 Transport_and_binding 0.834 2.033 # Enzyme/nonenzyme Prob Odds Enzyme 0.209 0.729 Nonenzyme => 0.791 1.109 # Enzyme class Prob Odds Oxidoreductase (EC 1.-.-.-) 0.056 0.268 Transferase (EC 2.-.-.-) 0.031 0.091 Hydrolase (EC 3.-.-.-) 0.062 0.195 Lyase (EC 4.-.-.-) 0.020 0.430 Isomerase (EC 5.-.-.-) 0.010 0.321 Ligase (EC 6.-.-.-) 0.017 0.327 # Gene Ontology category Prob Odds Signal_transducer 0.374 1.746 Receptor 0.128 0.750 Hormone => 0.247 37.936 Structural_protein 0.001 0.041 Transporter 0.025 0.228 Ion_channel 0.010 0.168 Voltage-gated_ion_channel 0.003 0.131 Cation_channel 0.010 0.215 Transcription 0.054 0.425 Transcription_regulation 0.091 0.724 Stress_response 0.099 1.128 Immune_response 0.178 2.090 Growth_factor 0.061 4.379 Metal_ion_transport 0.009 0.020 //
"Hormone" corresponds to GO:0005179 which is annotated in http://www.ebi.ac.uk/QuickGO/GProtein?ac=Q9Y5Q6 and due to that considered as true.
LAMP 1
############## ProtFun 2.2 predictions ############## >sp_P11279_L # Functional category Prob Odds Amino_acid_biosynthesis 0.011 0.484 Biosynthesis_of_cofactors 0.053 0.735 Cell_envelope => 0.804 13.186 Cellular_processes 0.027 0.373 Central_intermediary_metabolism 0.138 2.188 Energy_metabolism 0.037 0.411 Fatty_acid_metabolism 0.016 1.265 Purines_and_pyrimidines 0.533 2.195 Regulatory_functions 0.015 0.090 Replication_and_transcription 0.019 0.073 Translation 0.027 0.613 Transport_and_binding 0.834 2.033 # Enzyme/nonenzyme Prob Odds Enzyme 0.276 0.965 Nonenzyme => 0.724 1.014 # Enzyme class Prob Odds Oxidoreductase (EC 1.-.-.-) 0.039 0.187 Transferase (EC 2.-.-.-) 0.046 0.134 Hydrolase (EC 3.-.-.-) 0.058 0.184 Lyase (EC 4.-.-.-) 0.020 0.430 Isomerase (EC 5.-.-.-) 0.010 0.321 Ligase (EC 6.-.-.-) 0.017 0.326 # Gene Ontology category Prob Odds Signal_transducer 0.396 1.849 Receptor 0.282 1.659 Hormone 0.001 0.206 Structural_protein 0.011 0.408 Transporter 0.024 0.222 Ion_channel 0.008 0.147 Voltage-gated_ion_channel 0.002 0.111 Cation_channel 0.010 0.215 Transcription 0.032 0.247 Transcription_regulation 0.018 0.142 Stress_response 0.246 2.795 Immune_response => 0.371 4.368 Growth_factor 0.013 0.956 Metal_ion_transport 0.009 0.020 //
"Immune responds" corresponds to GO:0006955 which is not annotated in http://www.ebi.ac.uk/QuickGO/GProtein?ac=P11279 .
RET 4
############## ProtFun 2.2 predictions ############## >sp_P02753_R # Functional category Prob Odds Amino_acid_biosynthesis 0.017 0.751 Biosynthesis_of_cofactors 0.044 0.610 Cell_envelope => 0.804 13.186 Cellular_processes 0.075 1.021 Central_intermediary_metabolism 0.197 3.128 Energy_metabolism 0.043 0.475 Fatty_acid_metabolism 0.016 1.265 Purines_and_pyrimidines 0.275 1.131 Regulatory_functions 0.013 0.080 Replication_and_transcription 0.022 0.084 Translation 0.032 0.721 Transport_and_binding 0.800 1.951 # Enzyme/nonenzyme Prob Odds Enzyme => 0.544 1.900 Nonenzyme 0.456 0.639 # Enzyme class Prob Odds Oxidoreductase (EC 1.-.-.-) 0.095 0.458 Transferase (EC 2.-.-.-) 0.038 0.109 Hydrolase (EC 3.-.-.-) 0.235 0.742 Lyase (EC 4.-.-.-) => 0.059 1.264 Isomerase (EC 5.-.-.-) 0.010 0.321 Ligase (EC 6.-.-.-) 0.017 0.326 # Gene Ontology category Prob Odds Signal_transducer 0.202 0.942 Receptor 0.147 0.862 Hormone 0.004 0.667 Structural_protein 0.002 0.058 Transporter 0.025 0.232 Ion_channel 0.016 0.288 Voltage-gated_ion_channel 0.003 0.148 Cation_channel 0.010 0.215 Transcription 0.027 0.207 Transcription_regulation 0.025 0.196 Stress_response 0.161 1.829 Immune_response => 0.239 2.813 Growth_factor 0.023 1.617 Metal_ion_transport 0.009 0.020 //
"Immune responds" corresponds to GO:0006955 which is not annotated in http://www.ebi.ac.uk/QuickGO/GProtein?ac=P02753 .
Discussion
GOPET clearly outperformed the other methods for the GO-Term prediction. Its results were really detailled and mostly correct, while the results of Pfam and ProtFun were rather very generalized terms than detailed functions. However, also Pfam yielded at least a general hint on what the protein does. Contrary, the ProtFun were mostly wrong and so general, that they are rather useless in predciting the real function of the protein. Furthermore, we could not map the terms to real GO-identifier.
References
<references />
