Difference between revisions of "Sequence-based predictions"

From Bioinformatikpedia
(→‎OCTOPUS and SPOCTOPUS)
m (→‎General)
Line 382: Line 382:
   
 
We use the entries at UniProt for the real groundtruth and compare the prediction results shortly with them.
  
We use the entries at UniProt for the real groundtruth and compare the prediction results shortly with them.
  +
  +
  +
Why is the prediction of transmembrane helices and signal peptides grouped together here?
  +
Transmembrane helices and signal peptides are very hard to differ, if the method only predicts one of them. Therefore many false-positives are produced, because both of them consist almost only out of hydrophobic residues. Because of that, many methods gone the way, to combine the prediction of transmembrane helices and signal peptides, reducing drastically the false-positives.
  +
  +
There are different types of signal peptides, but they all work as an import or export information.
  +
* Import into the peroxisome
  +
* Import into the nucleus
  +
* Export from the nucleus
  +
* Import into the mitochondrium
   
 
==TMHMM==
  
==TMHMM==

Revision as of 14:31, 11 June 2011

Secondary structure prediction

PSIPRED

Secondary Structure predicted by PSIPRED
Source: http://bioinf.cs.ucl.ac.uk/psipred/

PSI-PRED use the PSI-BLAST output as input for a neuronal network which has a single hidden layer and a feed-forward back-propagation architecture to predict the secondary structure.

Results
PSI-PRED predicts a alpha/beta structure. The transmembrane region is predicted as a beta region. 

PSIPRED HFORMAT (PSIPRED V3.0)
Conf: 999851589999999877513567886245556456636899750389988756755687
Pred: CCCCCHHHHHHHHHHHHHHHCCCCCCCEEEEEEEEEEECCCCCCCEEEEEEEECCEEEEE
  AA: MGPRARPALLLLMLLQTAVLQGRLLRSHSLHYLFMGASEQDLGLSLFEALGYVDDQLFVF
             10        20        30        40        50        60
Conf: 318998225536664688990669998865311211002358577441156788603899
Pred: ECCCCCCEEECCCCCCCCCCHHHHHHHHHHHHCCCCCHHHHHHHHHHHCCCCCCCCEEEE
  AA: YDHESRRVEPRTPWVSSRISSQMWLQLSQSLKGWDHMFTVDFWTIMENHNHSKESHTLQV
             70        80        90       100       110       120
Conf: 987799319835459889765910588728988756689786135787788899999876
Pred: EEEEEEECCCEEEEEEEEEECCCEEEEECCCCCCCCCCCCCCHHHHHHHHHHHHHHHHHH
  AA: ILGCEMQEDNSTEGYWKYGYDGQDHLEFCPDTLDWRAAEPRAWPTKLEWERHKIRARQNR
            130       140       150       160       170       180
Conf: 310271499889888616322000378810000468999601699981450765189996
Pred: HHHCCCHHHHHHHHHHCCCCCCCCCCCCCEEEECCCCCCCEEEEEEEEEECCCCEEEEEE
  AA: AYLERDCPAQLQQLLELGRGVLDQQVPPLVKVTHHVTSSVTTLRCRALNYYPQNITMKWL
            190       200       210       220       230       240
Conf: 288106667520025355899875899999965999872169986699998826885259
Pred: ECCEECCCCCCCCCCCEECCCCCEEEEEEEEECCCCCCCEEEEEECCCCCCCEEEEEECC
  AA: KDKQPMDAKEFEPKDVLPNGDGTYQGWITLAVPPGEEQRYTCQVEHPGLDQPLIVIWEPS
            250       260       270       280       290       300
Conf: 999711124320001367777622367764115889887620212359
Pred: CCCCCEEEEEEEEEEEEEEEEEEEEEEEEEECCCCCCCCCCEEECCCC
  AA: PSGTLVIGVISGIAVFVVILFIGILFIILRKRQGSRGAMGHYVLAERE
            310       320       330       340

Jpred3

Jpred use the Jnet algorithm which provides "a three-state (a-helix, ß-strand and coil) prediction of secondary structure at an accuracy of 81.5%" <ref>http://nar.oxfordjournals.org/content/36/suppl_2/W197.abstract</ref>.

Results
Jpred found in it's first blast search a lot of homologous hits with an e-value range from e-163 to 4e-44. There are some self hits included. We continued to the prediction which is: 

Seq: MGPRARPALLLLMLLQTAVLQGRLLRSHSLHYLFMGASEQDLGLSLFEALGYVDD
 SS: ------HHHHHHHHHHHHH---------EEEEEEEEE-------EEEEEEEEE--

Seq: QLFVFYDHESRRVEPRTPWVSSRISSQMWLQLSQSLKGWDHMFTVDFWTIMENHN
 SS: EEEEEE-----EEEE----------HHHHHHHHHHHHHHHHHHHHHHHHHH----

Seq: HSKESHTLQVILGCEMQEDNSTEGYWKYGYDGQDHLEFCPDTLDWRAAEPRAWPT
 SS: -----EEEEEEEEEE------EEEEEEE-----EEEEEE----EEE-------HH

Seq: KLEWERHKIRARQNRAYLERDCPAQLQQLLELGRGVLDQQVPPLVKVTHHVTSSV
 SS: HHHHH--HHHHHHHHHH------HHHHHHHHHH-H-------EEEEE--------

Seq: TTLRCRALNYYPQNITMKWLKDKQPMDAKEFEPKDVLPNGDGTYQGWITLAVPPG
 SS: -EEEEEEE------EEEEEEE----------EE----------EEEEEEEEE---

Seq: EEQRYTCQVEHPGLDQPLIVIWEPSPSGTLVIGVISGIAVFVVILFIGILFIILR
 SS: ---EEEEEEEE------EEEEE---------HHHHHHHHHHHHHHHHHHHHHHHH
Seq: KRQGSRGAMGHYVLAERE
 SS: HH----------------

Comparison with DSSP

DSSP was designed by Wolfgang Kabsch and Chris Sander to provide a standard for the secondary structure assignment. DSSP calculates the secondary structure from PDB structures by using the distances between the atoms.

Results
Because, the PDB sequence is not complete, the dssp assignment is also incomplete. The interessting parts - the signal peptide and the cytoplasmic part - which are predicted as disordered are not covered by DSSP. PSIPRED and JPred predicted the transmembrane region well but assigned the - as disordered predicted - N- and C-terminus as a helical or beta sheet region. But the UniProt assignment gives no structure to this regions as well. Therefore, these regions may unstructured and not yet recognized as disordered regions. 

UniProt: ---------------------------EEEEEEEEEEE----EEE--EEEEEE--EEEEE
   DSSP:                          --EEEEEEEEEEB-SS-SSB--EEEEEETTEEEEE
PSIPRED: CCCCCHHHHHHHHHHHHHHHCCCCCCCEEEEEEEEEEECCCCCCCEEEEEEEECCEEEEE
  JPred: ------HHHHHHHHHHHHH---------EEEEEEEEE-------EEEEEEEEE--EEEEE
     AA: MGPRARPALLLLMLLQTAVLQGRLLRSHSLHYLFMGASEQDLGLSLFEALGYVDDQLFVF
DSSPSeq:                          RSHSLHYLFMGASEQDLGLSLFEALGYVDDQLFVF
                10        20        30        40        50        60
UniProt: EEEEE--EEE--------TTTHHHHHHHHHHHHHHHHHHHHHHHHHHTTT-EEE--EEEE
   DSSP: EESSS--EEE-STTS-SSTTTTHHHHHHHHHHHHHHHHHHHHHHHHHTTT-SSS--EEEE
PSIPRED: ECCCCCCEEECCCCCCCCCCHHHHHHHHHHHHCCCCCHHHHHHHHHHHCCCCCCCCEEEE
  JPred: E-----EEEE----------HHHHHHHHHHHHHHHHHHHHHHHHHH---------EEEEE
     AA: YDHESRRVEPRTPWVSSRISSQMWLQLSQSLKGWDHMFTVDFWTIMENHNHSKESHTLQV
DSSPSEQ: YDHESRRVEPRTPWVSSRISSQMWLQLSQSLKGWDHMFTVDFWTIMENHNHSKESHTLQV
                70        80        90       100       110       120
UniProt: EEEEEE-----EEEEEEEEE--EEEEEEEHHH-EEEEEE---HHHHHHHH---HHHHHHH
   DSSP: EEEEEE-TTS-EEEEEEEEETTEEEEEEEGGGTEEEESSGGGHHHHHHHHSSTHHHHHHH
PSIPRED: EEEEEEECCCEEEEEEEEEECCCEEEEECCCCCCCCCCCCCCHHHHHHHHHHHHHHHHHH
  JPred: EEEEE------EEEEEEE-----EEEEEE----EEE-------HHHHHHH--HHHHHHHH
     AA: ILGCEMQEDNSTEGYWKYGYDGQDHLEFCPDTLDWRAAEPRAWPTKLEWERHKIRARQNR
DSSPSEQ: ILGaEMQEDNSTEGYWKYGYDGQDHLEFCPDTLDWRAAEPRAWPTKLEWERHKIRARQNR
               130       140       150       160       170       180
UniProt: HHHH-HHHHHHHHHHHHHTTT-------EEEEEEEE----EEEEEEEEEEEEE--EEEEE
   DSSP: HHHHTHHHHHHHHHHHHHTTTSS--B--EEEEEEEE-SS-EEEEEEEEEEBSS--EEEEE
PSIPRED: HHHCCCHHHHHHHHHHCCCCCCCCCCCCCEEEECCCCCCCEEEEEEEEEECCCCEEEEEE
  JPred: HH------HHHHHHHHHH-H-------EEEEE---------EEEEEEE------EEEEEE
     AA: AYLERDCPAQLQQLLELGRGVLDQQVPPLVKVTHHVTSSVTTLRCRALNYYPQNITMKWL
DSSPSEQ: AYLERDaPAQLQQLLELGRGVLDQQVPPLVKVTHHVTSSVTTLRbRALNYYPQNITMKWL
               190       200       210       220       230       240
UniProt: E------HHH----EEEE-----EEEEEEEEE---HHHHEEEEEE---EEE-EEEE----
   DSSP: ETTEE--GGGS---EEEE-TTS-EEEEEEEEE-TTGGGGEEEEEE-TTSSS-EEEE-
PSIPRED: ECCEECCCCCCCCCCCEECCCCCEEEEEEEEECCCCCCCEEEEEECCCCCCCEEEEEECC
  JPred: E----------EE----------EEEEEEEEE------EEEEEEEE------EEEEE---
     AA: KDKQPMDAKEFEPKDVLPNGDGTYQGWITLAVPPGEEQRYTCQVEHPGLDQPLIVIWEPS
DSSPSEQ: KDKQPMDAKEFEPKDVLPNGDGTYQGWITLAVPPGEEQRYTbQVEHPGLDQPLIVIW
               250       260       270       280       290       300
UniProt: ------------------------------------------------
   DSSP:
PSIPRED: CCCCCEEEEEEEEEEEEEEEEEEEEEEEEEECCCCCCCCCCEEECCCC
  JPred: ------HHHHHHHHHHHHHHHHHHHHHHHHHH----------------
     AA: PSGTLVIGVISGIAVFVVILFIGILFIILRKRQGSRGAMGHYVLAERE
DSSPSEQ: 
               310       320       330       340

Prediction of disordered regions

The HFE-Gen is not yet known as disordered. It is not contained in the Disprot<ref>http://www.disprot.org/</ref> database. The prediction of unstructured regions predict serveral disorered regions in the protein, but most of them are predicted within secondary structure elements. Just the predicted disordered regions at the C- and N-terminus might be really unstructured but not yet experimentally recongnized because, these regions have no structural assignment.

The predictions are shown below.

DISOPRED

For the prediction, we used the DISOPRED-Server at http://bioinf.cs.ucl.ac.uk/disopred/
DISOPRED is a prediction tool for disordered regions based on a linear SVM. The SVM is trained with 750 non-redundant sequences with high resolution X-ray structures. "Disorder was identified with those residues that appear in the sequence records but with coordinates missing from the electron density map." <ref>http://bioinf.cs.ucl.ac.uk/index.php?id=806</ref> For each protein, a sequence profile was generated by using PSI-BLAST search against a filtered database. The PSI-BLAST profiles were used as input vectors for the SVM.

Result
Disopred predictes two disordered residues at the signal peptide and a disordered region at the end of the sequence which is located inside the cell.

DISOPRED prediction profile for the HFE protein
Source: http://bioinf.cs.ucl.ac.uk/disopred/
AA:Target sequence
Pred:Residue disorder prediction(.)= ordered residue(*)=Disordered residue
conf:997600000000000000000000000000000000000000000000000000000000
pred:**..........................................................
  AA:MGPRARPALLLLMLLQTAVLQGRLLRSHSLHYLFMGASEQDLGLSLFEALGYVDDQLFVF
             10        20	  30 	    40	      50	60
conf:000120011000000000000000000000000000000000000000000000000000
pred:............................................................
  AA:YDHESRRVEPRTPWVSSRISSQMWLQLSQSLKGWDHMFTVDFWTIMENHNHSKESHTLQV
             70        80	  90	   100	     110       120
conf:000000000000000000000000000000000000000000000000000000000000
pred:............................................................
  AA:ILGCEMQEDNSTEGYWKYGYDGQDHLEFCPDTLDWRAAEPRAWPTKLEWERHKIRARQNR
            130       140       150       160       170       180
conf:000000000000000000000002456777878777766530000000000000000000
pred:..............................*.*...........................
  AA:AYLERDCPAQLQQLLELGRGVLDQQVPPLVKVTHHVTSSVTTLRCRALNYYPQNITMKWL
            190       200       210       220       230       240
conf:000035555545543000000000000000000000000000000000000001354667
pred:............................................................
  AA:KDKQPMDAKEFEPKDVLPNGDGTYQGWITLAVPPGEEQRYTCQVEHPGLDQPLIVIWEPS
            250       260       270       280       290       300
conf:777766643300000000000000047889999999999999898999
pred:...........................*********************
  AA:PSGTLVIGVISGIAVFVVILFIGILFIILRKRQGSRGAMGHYVLAERE
            310       320       330       340
DISOPRED predictions for a false positive rate threshold of: 2%

POODLE

POODLE stands for Prediction Of Order and Disorder by machine LEarning.

POODLE provides three different predictions

  • POODLE-S: short disorder regions prediction
  • POODLE-L: long disorder regions prediction (longer 40 residues)
  • unfolded protein prediction


All POODLE variants predict a disordered region at the end of the protein which contains a transmembrane region (pos: 307-330), this shows an evidance for a disordered region at the C-Terminus. But also, all variants predict a short disordered region at the beginning of the sequence which is a part of the signal peptid (pos: 1-22).

POODLE-I

POODLE-I (series only) predicted 4 disordered regions within the protein sequence.

Distribution of disordered region over the AS-Sequence predicted by POODLE-I
Source: http://mbs.cbrc.jp/poodle/poodle.html
MGPRARPALLLLMLLQTAVLQGRLLRSHSLHYLFMGASEQDLGLSLFEALGYVDDQLFVF
**************----------------------------------------------
YDHESRRVEPRTPWVSSRISSQMWLQLSQSLKGWDHMFTVDFWTIMENHNHSKESHTLQV
-------**********---******------*---------------------------
ILGCEMQEDNSTEGYWKYGYDGQDHLEFCPDTLDWRAAEPRAWPTKLEWERHKIRARQNR
------------------------------------------------------------
AYLERDCPAQLQQLLELGRGVLDQQVPPLVKVTHHVTSSVTTLRCRALNYYPQNITMKWL
---------------------***************------------------------
KDKQPMDAKEFEPKDVLPNGDGTYQGWITLAVPPGEEQRYTCQVEHPGLDQPLIVIWEPS
----*********----------------------------------------*******
PSGTLVIGVISGIAVFVVILFIGILFIILRKRQGSRGAMGHYVLAERE
************************************************

POODLE-S

POODLE-S (using missing residues) predicts 6 short disordered regions within the protein sequence.

Distribution of disordered region over the AS-Sequence predicted by POODLE-S(Missing residues)
Source: http://mbs.cbrc.jp/poodle/poodle.html
MGPRARPALLLLMLLQTAVLQGRLLRSHSLHYLFMGASEQDLGLSLFEALGYVDDQLFVF
-**************---------------------------------------------
YDHESRRVEPRTPWVSSRISSQMWLQLSQSLKGWDHMFTVDFWTIMENHNHSKESHTLQV
-------**********---******----------------------------------
ILGCEMQEDNSTEGYWKYGYDGQDHLEFCPDTLDWRAAEPRAWPTKLEWERHKIRARQNR
------------------------------------------------------------
AYLERDCPAQLQQLLELGRGVLDQQVPPLVKVTHHVTSSVTTLRCRALNYYPQNITMKWL
---------------------***************------------------------
KDKQPMDAKEFEPKDVLPNGDGTYQGWITLAVPPGEEQRYTCQVEHPGLDQPLIVIWEPS
----*********----------------------------------------*******
PSGTLVIGVISGIAVFVVILFIGILFIILRKRQGSRGAMGHYVLAERE
*--------------------------------********-------

POODLE-S (using High B-Factor residues) predicts 2 short disordered regions within the protein sequence.

Distribution of disordered region over the AS-Sequence predicted by POODLE-S(High B-Factor residues)
Source: http://mbs.cbrc.jp/poodle/poodle.html
MGPRARPALLLLMLLQTAVLQGRLLRSHSLHYLFMGASEQDLGLSLFEALGYVDDQLFVF
-*-***------------------------------------------------------
YDHESRRVEPRTPWVSSRISSQMWLQLSQSLKGWDHMFTVDFWTIMENHNHSKESHTLQV
------------------------------------------------------------
ILGCEMQEDNSTEGYWKYGYDGQDHLEFCPDTLDWRAAEPRAWPTKLEWERHKIRARQNR
------------------------------------------------******------
AYLERDCPAQLQQLLELGRGVLDQQVPPLVKVTHHVTSSVTTLRCRALNYYPQNITMKWL
------------------------------------------------------------
KDKQPMDAKEFEPKDVLPNGDGTYQGWITLAVPPGEEQRYTCQVEHPGLDQPLIVIWEPS
------------------------------------------------------------
PSGTLVIGVISGIAVFVVILFIGILFIILRKRQGSRGAMGHYVLAERE
------------------------------------------------

POODLE-L

POODLE-L predicts a disordered region from 296 to the end.

Distribution of disordered region over the AS-Sequence predicted by POODLE-L
Source: http://mbs.cbrc.jp/poodle/poodle.html
MGPRARPALLLLMLLQTAVLQGRLLRSHSLHYLFMGASEQDLGLSLFEALGYVDDQLFVF 
------------------------------------------------------------
YDHESRRVEPRTPWVSSRISSQMWLQLSQSLKGWDHMFTVDFWTIMENHNHSKESHTLQV 
------------------------------------------------------------
ILGCEMQEDNSTEGYWKYGYDGQDHLEFCPDTLDWRAAEPRAWPTKLEWERHKIRARQNR 
------------------------------------------------------------
AYLERDCPAQLQQLLELGRGVLDQQVPPLVKVTHHVTSSVTTLRCRALNYYPQNITMKWL 
------------------------------------------------------------
KDKQPMDAKEFEPKDVLPNGDGTYQGWITLAVPPGEEQRYTCQVEHPGLDQPLIVIWEPS 
------------------------------------------------------******
PSGTLVIGVISGIAVFVVILFIGILFIILRKRQGSRGAMGHYVLAERE
************************************************

IUPRED

IUPRED use the estimated pairwise energy to recognize unstructured regions within protein sequnces. For these, they use the assumption, that all globular proteins have an amino acid composision which gives it the potential to form a large number of favorable interactions.

Results
The short term prediction predicts 5 short regions. There are also disordered residues at the beginning and in the signal peptide.

IUPRED prediction of short regions
Source: http://iupred.enzim.hu/
IUPRED prediction of long regions
Source: http://iupred.enzim.hu/
IUPRED prediction of structured regions
Source: http://iupred.enzim.hu/
MGPRARPALLLLMLLQTAVLQGRLLRSHSLHYLFMGASEQDLGLSLFEALGYVDDQLFVF 
***---------------------------------------------------------
YDHESRRVEPRTPWVSSRISSQMWLQLSQSLKGWDHMFTVDFWTIMENHNHSKESHTLQV 
------------------------------------------------------------
ILGCEMQEDNSTEGYWKYGYDGQDHLEFCPDTLDWRAAEPRAWPTKLEWERHKIRARQNR 
------------------------------------------------------------
AYLERDCPAQLQQLLELGRGVLDQQVPPLVKVTHHVTSSVTTLRCRALNYYPQNITMKWL 
------------------------------------------------------------
KDKQPMDAKEFEPKDVLPNGDGTYQGWITLAVPPGEEQRYTCQVEHPGLDQPLIVIWEPS 
---------********----------***--------*-****----------------
PSGTLVIGVISGIAVFVVILFIGILFIILRKRQGSRGAMGHYVLAERE
---------------------------------------------***


The long term prediction predicted 7 disordered residues, but just one short region. 

MGPRARPALLLLMLLQTAVLQGRLLRSHSLHYLFMGASEQDLGLSLFEALGYVDDQLFVF 
------------------------------------------------------------
YDHESRRVEPRTPWVSSRISSQMWLQLSQSLKGWDHMFTVDFWTIMENHNHSKESHTLQV 
------------------------------------------------------------
ILGCEMQEDNSTEGYWKYGYDGQDHLEFCPDTLDWRAAEPRAWPTKLEWERHKIRARQNR 
------------------------------------------------------------
AYLERDCPAQLQQLLELGRGVLDQQVPPLVKVTHHVTSSVTTLRCRALNYYPQNITMKWL 
------------------------------------------------------------
KDKQPMDAKEFEPKDVLPNGDGTYQGWITLAVPPGEEQRYTCQVEHPGLDQPLIVIWEPS 
---------******-------------------------*-------------------
PSGTLVIGVISGIAVFVVILFIGILFIILRKRQGSRGAMGHYVLAERE
------------------------------------------------




The prediction of sturcured regions predicts one globular domain from 1-348. This means, that the whole protein is structured. This is a contradiction to the prediction of POODLE, but because of the weak evidence given by the other IUPRED-methods not a real contradiction to the other results of IUPRED.

META-Disorder

For this task, we used the PredictProtein Server at https://www.predictprotein.org. META-Disorder, published in 2009 by Avner Schlessinger, Marco Punta, Guy Yachdav, Laszlo Kajan and Burkhard Rost, use a combined prediction of ORSnet PROFbval and Ucon.

predicted secondary structure composision

sec str type H E L
% in protein 27.30 28.74 43.97


Prediction of disordered residues by META-Disorder (last coloumn) 

Number Residue NORSnet NORS2st PROFbval bval2st Ucon Ucon2st MD_raw   MD_rel  MD2st 

                                 ....
 242	D	0.13	-	0.70	D	0.76	D	0.444	2	-
 243	K	0.13	-	0.69	D	0.76	D	0.480	1	-
 244	Q	0.13	-	0.66	D	0.93	D	0.531	0	D
 245	P	0.17	-	0.73	D	0.92	D	0.520	0	D
 246	M	0.28	-	0.65	D	0.90	D	0.525	0	D
 247	D	0.31	-	0.68	D	0.87	D	0.515	0	-
 248	A	0.36	-	0.70	D	0.87	D	0.520	0	D
 249	K	0.40	-	0.69	D	0.89	D	0.485	1	-
                                 ....
 344	L	0.37	-	0.59	D	0.18	-	0.515	0	-
 345	A	0.35	-	0.74	D	0.17	-	0.515	0	-
 346	E	0.31	-	0.89	D	0.17	-	0.520	0	D
 347	R	0.35	-	0.91	D	0.23	-	0.525	0	D
 348	E	0.34	-	0.92	D	0.38	-	0.520	0	D

Key for output
----------------
Number - residue number
Residue - amino-acid type
NORSnet - raw score by NORSnet (prediction of unstructured loops)
NORS2st - two-state prediction by NORSnet; D=disordered
PROFbval - raw score by PROFbval (prediction of residue flexibility from sequence)
Bval2st - two-state prediction by PROFbval
Ucon - raw score by Ucon (prediction of protein disorder using predicted internal contacts)
Ucon2st - two-state prediction by Ucon
MD - raw score by MD (prediction of protein disorder using orthogonal sources)
MD_rel - reliability of the prediction by MD; values range from 0-9. 9=strong prediction
MD2st - two-state prediction by MD

META-Disorder predict a very short disordered region of 3 residues at the end of the protein but with a week evidence of around 0.5. Therefore it is quite unlikely to have a disorered region at the C-Terminus if we just look at this method.

Prediction of transmembrane alpha-helices and signal peptides

General

We were given five additional proteins to work with and predict transmembrane regions, signal peptides and GO terms for. That was done, because most of the practials proteins are no membrane proteins and therefore produce only "no membrane" results. Thus the three membrane proteins [BACR_HALSA], [LAMP1_HUMAN] and [A4_HUMAN] were provided, but also our HFE Protein [HFE_HUMAN] is an membrane protein.

To give you a quick overview about the protein properties, look at the following table:

Accession Entry name Organism Subcelluar location Signal peptide
Q30201 HFE_HUMAN Homo sapiens (Human) Membrane; Single-pass type I membrane protein 1-22
P02945 BACR_HALSA Halobacterium salinarium / (Halobacterium halobium) Cell membrane; Multi-pass membrane protein no
P02753 RET4_HUMAN Homo sapiens (Human) Secreted 1-18
Q9Y5Q6 INSL5_HUMAN Homo sapiens (Human) Secreted 1-22
P11279 LAMP1_HUMAN Homo sapiens (Human) Cell membrane; Single-pass type I membrane protein [...] 1-28
P05067 A4_HUMAN Homo sapiens (Human) Membrane; Single-pass type I membrane protein 1-17

We are going to predict membranes and signaling for these six proteins using different tools. Because our normally adressed protein HFE_HUMAN is an membrane protein and therefore we see the prediction accurancy by using it, we will give only graphical and detailed overview about the results of HFE_HUMAN and group the additional proteins in textual form.

We use the entries at UniProt for the real groundtruth and compare the prediction results shortly with them.


Why is the prediction of transmembrane helices and signal peptides grouped together here? Transmembrane helices and signal peptides are very hard to differ, if the method only predicts one of them. Therefore many false-positives are produced, because both of them consist almost only out of hydrophobic residues. Because of that, many methods gone the way, to combine the prediction of transmembrane helices and signal peptides, reducing drastically the false-positives.

There are different types of signal peptides, but they all work as an import or export information.

  • Import into the peroxisome
  • Import into the nucleus
  • Export from the nucleus
  • Import into the mitochondrium

TMHMM

TMHMM posterior probabilities

TMHMM is a tool for predicting membrane topology (transmembrane helices) in proteins based on a hidden Markov model with different states. It devides the regions in "inside", "outside" and "TMhelix". But TMHMM can not predict Signal Peptides

TMHMM was used locally in our linux box, after correcting some path issues inside some config files.

The command we used was:

  • tmhmm x.fasta > x.tmhmm

where 'x' stands for one of the UniProt entry name of the proteins. Afterwards we tried to plot the result of HFE_HUMAN with gnuplot, but this was not working either, because of path issues inside the (automatically created) gnuplot script of tmhmm. After correcting the path issues again, gnuplot worked fine and produced successfully graphical output.

A color code is applied to provide easier reading:

  • green: The predicted region matches the experimental resolved region from UniProt (+/- 5 residues allowed). TMHMM succeeded.
  • yellow: The predicted region does only partially match the experimental region (mostly errors with signaling peptides). TMHMM has to be improved.
  • red: The predicted region does absolutely not match the experimental region. TMHMM failed.


TMHMM UniProt
id version region start end region start end
Q30201|HFE_HUMAN TMHMM2.0 outside 1 306 Signal peptide 1 22
Q30201|HFE_HUMAN Extracelluar 23 306
Q30201|HFE_HUMAN TMHMM2.0 TMhelix 307 329 Helical 307 330
Q30201|HFE_HUMAN TMHMM2.0 inside 330 348 Cytoplasmic 331 348

TMHMM misses clearly the the signal peptide and counts the region as outside (1-306), which is correct according to UniProt. Also the TMhelix (307-329) and the inside region (330-348) is placed right, only with one amino acid deviation, but that is insignificant. Therefore TMHMM was very successful in prediction of right regions. The results are shown in the figure to the right, too.

TMHMM UniProt
id version region start end region start end
sp_P02945_BACR_HALSA TMHMM2.0 outside 1 22 Extracellular 14 23
sp_P02945_BACR_HALSA TMHMM2.0 TMhelix 23 42 Helical; Name=Helix A 24 42
sp_P02945_BACR_HALSA TMHMM2.0 inside 43 54 Cytoplasmic 43 56
sp_P02945_BACR_HALSA TMHMM2.0 TMhelix 55 77 Helical; Name=Helix B 57 75
sp_P02945_BACR_HALSA TMHMM2.0 outside 78 91 Extracellular 76 91
sp_P02945_BACR_HALSA TMHMM2.0 TMhelix 92 114 Helical; Name=Helix C 92 109
sp_P02945_BACR_HALSA TMHMM2.0 inside 115 120 Cytoplasmic 110 120
sp_P02945_BACR_HALSA TMHMM2.0 TMhelix 121 143 Helical; Name=Helix D 121 140
sp_P02945_BACR_HALSA TMHMM2.0 outside 144 147 Extracellular 141 147
sp_P02945_BACR_HALSA TMHMM2.0 TMhelix 148 170 Helical; Name=Helix E 148 167
sp_P02945_BACR_HALSA TMHMM2.0 inside 171 189 Cytoplasmic 168 185
sp_P02945_BACR_HALSA TMHMM2.0 TMhelix 190 212 Helical; Name=Helix F 186 204
sp_P02945_BACR_HALSA TMHMM2.0 outside 213 262 Extracellular 205 216
Helical; Name=Helix G 217 236
Cytoplasmic 237 262

As clearly visible, TMHMM is also capable of prediction regions of non eukaroynts almost completely correctly. Only the last helical and cytoplasmic regions are missed.

TMHMM UniProt
id version region start end region start end
sp_P02753_RET4_HUMAN TMHMM2.0 outside 1 201 Signal peptide 1 18

TMHMM misses the signaling peptide but everything else is superb, because RET4_Human is a protein that gets secreted.

TMHMM UniProt
id version region start end region start end
sp_Q9Y5Q6_INSL5_HUMAN TMHMM2.0 outside 1 135 Signal peptide 1 22

As usual, the TMHMM misses the signaling peptide but predicts accuratly. INSL5_Human is a hormone and thus sectred at the extracellular regions.

TMHMM UniProt
id version region start end region start end
sp_P11279_LAMP1_HUMAN TMHMM2.0 inside 1 10 Signal peptide 1 28
sp_P11279_LAMP1_HUMAN TMHMM2.0 TMhelix 11 33
sp_P11279_LAMP1_HUMAN TMHMM2.0 outside 34 383 Lumenal 29 382
sp_P11279_LAMP1_HUMAN TMHMM2.0 TMhelix 384 406 Helical 383 405
sp_P11279_LAMP1_HUMAN TMHMM2.0 inside 407 417 Cytoplasmic 406 417

TMHMM does not so well predict this protein. It mixes up the region of the signaling peptide with 'inside' and 'TMhelix' and is completely wrong at lumenal with 'outside'. The rest is predicted well.

TMHMM UniProt
id version region start end region start end
sp_P05067_A4_HUMAN TMHMM2.0 outside 1 700 Signal peptide 1 17
Extracellular 18 699
sp_P05067_A4_HUMAN TMHMM2.0 TMhelix 701 723 Helical 700 723
sp_P05067_A4_HUMAN TMHMM2.0 inside 724 770 Cytoplasmic 724 770

TMHMM predicted again almost correctly and missed only the beginning unpredictable signaling peptide.

Phobius and PolyPhobius

For Phobius and PolyPhobius, we used the webservice<ref>http://www.ncbi.nlm.nih.gov/pubmed/17483518?dopt=Abstract</ref> at http://phobius.sbc.su.se/ with standard settings.

Phobius is a combined predictor for transmembrane protein topology and signal peptide. Phobius models different regions of the seuqence in a series of interconnected states of a HMM.<ref>http://www.ncbi.nlm.nih.gov/pubmed/15111065?dopt=Abstract</ref>
PolyPhobius is a hidden Markov model (HMM) decoding algorithm. It combines probabilities for sequence features of homologs by considering the average of the posterior label probability of each position in a global sequence alignment. PolyPhobius is benchmarked by Phobius. <ref>http://www.ncbi.nlm.nih.gov/pubmed/15961464?dopt=Abstract</ref>

Phobius

predicted regions by Phobius
Source: http://phobius.sbc.su.se/

Phobius predicts very accurate as seen below. The transmembrane region is predicted just 1-2 residues upstream from the annotated region. The same holds for the topological domains before and after the transmembrane region. Also the signal peptid is correctly predicted. 

PREDICTED                                                     ANNOTATION
ID   sp|Q30201|HFE_HUMAN
FT   SIGNAL        1     21                             |  1-20
FT   REGION        1      7       N-REGION.              
FT   REGION        8     16       H-REGION.
FT   REGION       17     21       C-REGION.
FT   TOPO_DOM     22    304       NON CYTOPLASMIC.      |  23-306
FT   TRANSMEM    305    329                             |  307-330
FT   TOPO_DOM    330    348       CYTOPLASMIC.          |  331-348

PolyPhobius

predicted regions by PolyPhobius
Source: http://phobius.sbc.su.se/

PolyPhobius also predicts very accurate but in our case not as accurate as Phobius. 

PREDICTED                                                     ANNOTATION
ID   sp|Q30201|HFE_HUMAN
FT   SIGNAL        1     23                             |  1-20
FT   REGION        1      5       N-REGION.              
FT   REGION        6     19       H-REGION.
FT   REGION       20     23       C-REGION.
FT   TOPO_DOM     24    304       NON CYTOPLASMIC.      |  23-306
FT   TRANSMEM    305    329                             |  307-330
FT   TOPO_DOM    330    348       CYTOPLASMIC.          |  331-348


Additional proteins

  • BACR_HALSA: Phobius/Polyphobius are almost the same. There is only a slight change in the domain length, furthermore both methods predicted the membrane topologies right.
  • RET4_HUMAN: Phobius/Polyphobius predicted the position of the signaling peptide correct, the overall prediction is correct, too.
  • INSL5_HUMAN: Phobius/Polyphobius predict again verry accurate and correct. The signaling peptide and extracelluar region are at the correct positions.
  • LAMP1_HUMAN: Phobius/Polyphobius predicted the membrane topology correctly.
  • A4_HUMAN: Phobius/Polyphobius predicted signaling peptide position is correct.

Summing up, Phobius/Polyphobius are very accurate at their predictions about the membrane topology. Polyphobius takes much more time to produce an result, but provide no really better result - sometimes even a bit worser. Because of that, we decided that Phobious is the more comfortable one.

OCTOPUS and SPOCTOPUS

OCTOPUS is a combined mehtod of HMM's and artificial neural networks. OCTOPUS first create a sequence profile by homology search using BLAST. The profile is used as the input to a set of neural networks which predict the preferance of the location for each residue. Each residue is predicted to be either inside or outside the cell and located in a transmembrane (M), interface (I), close loop (L) or globular loop (G) environment.
SPOCTOPUS is an extended version of OCTOPUS that can also predict signal peptides. It use a neural network to predict a signal peptide if the score for each of the 70 N-Terminal residues is high enough.


Both, OCTOPUS and SPOCTOPUS predict the signal peptide and the transmembrane region correctly as you can see in the images below. Also both methods predict a signal peptide at the N-terminus which has the correct length.

predicted regions by OCTOPUS
Source: http://octopus.cbr.su.se/
predicted regions by SPOCTOPUS
Source: http://octopus.cbr.su.se/

Additional proteins


BACR_HALSA by left: OCTO- right: SPOCTOPUS, source: http://OCTOPUS.cbr.su.se/index.php
RET4_HUMAN by left: OCTO- right: SPOCTOPUS, source: http://OCTOPUS.cbr.su.se/index.php
INSL5_HUMAN by left: OCTO- right: SPOCTOPUS, source: http://OCTOPUS.cbr.su.se/index.php
LAMP1_HUMAN by left: OCTO- right: SPOCTOPUS, source: http://OCTOPUS.cbr.su.se/index.php
A4_HUMAN by left: OCTO- right: SPOCTOPUS, source: http://OCTOPUS.cbr.su.se/index.php
  • BACR_HALSA:

The prediction result of OCTOPUS and SPOCTOPUS is the same. They are both correct, because BACR_HALSA does not contain any signal peptides.

  • RET4_HUMAN:

OCTOPUS predictes an TM-helix instead of the signal peptide, SPOCTOPUS corrects this error.

  • INSL5_HUMAN:

Same error for OCTOPUS as seen by RET4_HUMAN, SPOCTOPUS corrects this error again.

  • LAMP1_HUMAN:

OCTOPUS and SPOCTOPUS predict both the TM-helix correctly. But same error for OCTOPUS as seen with RET4_HUMAN, INSL5_HUMAN: instead of signaling peptide an TM-helix and SPOCTOPUS corrects that.

  • A4_HUMAN:

OCTOPUS predictes an Reentrant/Dip region instead of the signal peptide region, SPOCTOPUS corrects that.

SignalP

For using it locally at our linux box, we had to correct again some path issues.

The command we used was:

  • signalp -t y x.fasta > x.signalp

where 'x' stands again for the UniProt entry names of the proteins. 'y' was chosen accordingly to the organism of the protein, for all human proteins 'y' was set to eukaryotes 'euk' and for the bacterial protein P02945 to gram- 'gram-'. This switch specifies the neural network and hidden Markov models, that are seperatly trained for different organismns.

For the graphical output of HFE_HUMAN we used the SignalP server from: http://www.cbs.dtu.dk/services/SignalP

There are three scorings for the SignalP-prediction NN:

  • C-score: 'cleavage site': raw cleavage site prediction
  • S-mean-score: 'average of the S-score': discrimination of secretory and non-secretory proteins
  • Y-max-score: 'combination of C-score with s-core': better cleavage site prediction
SignalP-NN prediction (source: signalp)
SignalP-NN result:
sp|Q30201|HFE_HUMAN   length = 348
Measure   Position Value   Cutoff signal peptide?
max. C    23       0.534   0.32   YES
max. Y    23       0.599   0.33   YES
max. S    16       0.995   0.87   YES
mean S    1-22     0.935   0.48   YES
     D    1-22     0.767   0.43   YES
Most likely cleavage site between pos. 22 and 23: LQG-RL
SignalP-HMM prediction (source: signalp)


SignalP-HMM result:
>sp|Q30201|HFE_HUMAN
Prediction: Signal peptide
Signal peptide probability: 0.998
Signal anchor probability: 0.000
Max cleavage site probability: 0.297 between pos. 22 and 23

SignalP predicts an signal peptide probability with almost 1.0 and thus an signal anchor probability with 0. This leads to the prediction of an cleavage site between pos. 22 and 23.

According to UniProt is there an signal peptide, it starts at pos. 1 to 22, which means, SignalP has predicted the signal peptide and cleavage site with 100% accurancy. 


SignalP-NN result:
>sp_P02945_BACR_HALSA  length = 70
Measure  Position  Value  Cutoff  signal peptide?
max. C    16       0.331   0.52   NO
max. Y    43       0.066   0.33   NO
max. S    32       0.948   0.92   YES
mean S     1-42    0.216   0.49   NO
     D     1-42    0.141   0.44   NO
Most likely cleavage site between pos. 42 and 43: FLV-KG

SignalP-HMM result:
>sp|P02945|BACR_HALSA
Prediction: Non-secretory protein
Signal peptide probability: 0.000
Max cleavage site probability: 0.000 between pos. 15 and 16

For BACR_Halsa is the result of SignalP clearly wrong, because this bacteria does not contain any signal peptide. 


SignalP-NN result:
>sp_P02753_RET4_HUMAN  length = 70
Measure  Position  Value  Cutoff  signal peptide?
max. C    19       0.929   0.32   YES
max. Y    19       0.901   0.33   YES
max. S     1       0.994   0.87   YES
mean S     1-18    0.938   0.48   YES
     D     1-18    0.920   0.43   YES
Most likely cleavage site between pos. 18 and 19: GRA-ER

SignalP-HMM result:
>sp_P02753_RET4_HUMAN
Prediction: Signal peptide
Signal peptide probability: 1.000
Signal anchor probability: 0.000
Max cleavage site probability: 0.979 between pos. 18 and 19

SignalP predicted the cleavage very well, accoring to UniProt the signal peptide is from pos 1 to 18 and afterwards the cleavage. 


SignalP-NN result:
>sp_Q9Y5Q6_INSL5_HUMA  length = 70
Measure  Position  Value  Cutoff  signal peptide?
max. C    23       0.855   0.32   YES
max. Y    23       0.778   0.33   YES
max. S    13       0.987   0.87   YES
mean S     1-22    0.852   0.48   YES
     D     1-22    0.815   0.43   YES
Most likely cleavage site between pos. 22 and 23: VRS-KE

SignalP-HMM result:
>sp_Q9Y5Q6_INSL5_HUMAN
Prediction: Signal peptide
Signal peptide probability: 0.999
Signal anchor probability: 0.000
Max cleavage site probability: 0.911 between pos. 22 and 23

This result is also correct predicted (UniProt: signal peptide from pos 1-22). 

SignalP-NN result:
>sp_P11279_LAMP1_HUMA  length = 70
Measure  Position  Value  Cutoff  signal peptide?
max. C    29       0.978   0.32   YES
max. Y    29       0.903   0.33   YES
max. S    19       0.999   0.87   YES
mean S     1-28    0.960   0.48   YES
     D     1-28    0.932   0.43   YES
Most likely cleavage site between pos. 28 and 29: ASA-AM

SignalP-HMM result:
>sp_P11279_LAMP1_HUMAN
Prediction: Signal peptide
Signal peptide probability: 1.000
Signal anchor probability: 0.000
Max cleavage site probability: 0.847 between pos. 28 and 29

SignalP predicted again correctly the cleavage site. 


SignalP-NN result:
>sp_P05067_A4_HUMAN    length = 70
Measure  Position  Value  Cutoff  signal peptide?
max. C    18       0.891   0.32   YES
max. Y    18       0.850   0.33   YES
max. S     2       0.992   0.87   YES
mean S     1-17    0.967   0.48   YES
     D     1-17    0.909   0.43   YES
Most likely cleavage site between pos. 17 and 18: ARA-LE

SignalP-HMM result:
>sp_P05067_A4_HUMAN
Prediction: Signal peptide
Signal peptide probability: 1.000
Signal anchor probability: 0.000
Max cleavage site probability: 0.993 between pos. 17 and 18

That prediction is accurate, too.

In short, SignalP predicts very accurate the position of the signaling peptide cleavage side, but fails with the non-eukaryont BACR_HALSA.

TargetP

TargetP predict for each of the proteins a signal preptide with high probability. But P02945 which is a bacteria and has no signal peptide, the method seems to be pretty accurate. It scores the prediction at the P02945 with an relibility clause of 4, which is almost neglectable (1 is the highest coinfidence, 5 is the worst). 

### targetp v1.1 prediction results ##################################
Number of query sequences:  6
Cleavage site predictions included.
Using NON-PLANT networks.
Name                  Len            mTP     SP  other  Loc  RC  TPlen
----------------------------------------------------------------------
sp_Q30201_HFE_HUMAN   348          0.433  0.912  0.004   S    3     22
sp_P02945_BACR_HALSA  262          0.019  0.897  0.562   S    4    116
sp_P02753_RET4_HUMAN  201          0.242  0.928  0.020   S    2     18
sp_Q9Y5Q6_INSL5_HUMA  135          0.074  0.899  0.037   S    1     22
sp_P11279_LAMP1_HUMA  417          0.043  0.953  0.017   S    1     28
sp_P05067_A4_HUMAN    770          0.035  0.937  0.084   S    1     17
----------------------------------------------------------------------
cutoff                             0.000  0.000  0.000

Prediction of GO terms

General

GO-Terms classify protein functions. Each GO-Term states other protein functions, therefore classifying a protein into GO-Terms means predicting it's functions.

HFE_HUMAN is annotated with 27 different GO Terms which are <ref>http://www.ebi.ac.uk/QuickGO/GProtein?ac=Q30201</ref>:

GOID GO Term Aspect
GO:0002474 antigen processing and presentation of peptide antigen via MHC class I Process
GO:0005515 protein binding Function
GO:0005737 cytoplasm Component
GO:0005769 early endosome Component
GO:0005886 plasma membrane Component
GO:0005887 integral to plasma membrane Component
GO:0006461 protein complex assembly Process
GO:0006810 transport Process
GO:0006811 ion transport Process
GO:0006826 iron ion transport Process
GO:0006879 cellular iron ion homeostasis Process
GO:0006898 receptor-mediated endocytosis Process
GO:0006955 immune response Process
GO:0007565 female pregnancy Process
GO:0010106 cellular response to iron ion starvation Process
GO:0016020 membrane Component
GO:0016021 integral to membrane Component
GO:0019882 antigen processing and presentation Process
GO:0031410 cytoplasmic vesicle Component
GO:0042446 hormone biosynthetic process Process
GO:0042612 MHC class I protein complex Component
GO:0045177 apical part of cell Component
GO:0045178 basal part of cell Component
GO:0048471 perinuclear region of cytoplasm Component
GO:0055037 recycling endosome Component
GO:0055072 iron ion homeostasis Process
GO:0060586 multicellular organismal iron ion homeostasis Process

GOPET

Gopet predicted 2 GO-Terms which have no overlap to the annotation.

GOID Aspect Confidence GO Term
GO:0004872 Molecular Function 91% receptor activity
GO:0030106 Molecular Function 88% MHC class I receptor activity

Pfam

Pfam is a database that contains protein domains and families. For our search we used the webserver at http://pfam.sanger.ac.uk/search with standard values.

Afterwards we used the pfam2go database, to find the GO-entries matching the pfam descriptions.

Pfam classifies the HFE_Human protein into two families:

Pfam classification of protein families (source: pfam)


  • Family: MHC_I (PF00129)
  • Family: C1-set (PF07654)

For the PF00129 family are four hits at the pfam2go data: 

Pfam:PF00129 MHC_I > GO:immune response ; GO:0006955
Pfam:PF00129 MHC_I > GO:antigen processing and presentation ; GO:0019882
Pfam:PF00129 MHC_I > GO:membrane ; GO:0016020
Pfam:PF00129 MHC_I > GO:MHC class I protein complex ; GO:0042612
Significant Pfam-A matches (source: pfam)

All those GO-Entries are at the UniProt entry about HFE_Human, so this family is correct.

For the PF07654 family are no entries at the pfam2go data and thus no validateable crosslinks to UniProt, maybe this family is yet not included in the pfam2go data.

For a more detailed picture have a look at the figure on the right, you can see the Pfam-A matches with alignment.

ProtFun 2.2

ProtFun is an ab initio prediction server.

Results
ProtFun assigned immune response(GO:0006955;Process) to HFE what is correct. But ProtFun predicts just one correct GO-number for the HFE-Gen. 

 Functional category                  Prob     Odds
 Amino_acid_biosynthesis              0.011    0.484
 Biosynthesis_of_cofactors            0.105    1.452
 Cell_envelope                     => 0.633   10.377
 Cellular_processes                   0.095    1.297
 Central_intermediary_metabolism      0.231    3.663
 Energy_metabolism                    0.059    0.659
 Fatty_acid_metabolism                0.016    1.265
 Purines_and_pyrimidines              0.583    2.400
 Regulatory_functions                 0.013    0.079
 Replication_and_transcription        0.019    0.073
 Translation                          0.079    1.801
 Transport_and_binding                0.732    1.785


 Enzyme/nonenzyme                     Prob     Odds
 Enzyme                               0.208    0.727
 Nonenzyme                         => 0.792    1.110


 Enzyme class                         Prob     Odds
 Oxidoreductase (EC 1.-.-.-)          0.084    0.404
 Transferase    (EC 2.-.-.-)          0.062    0.179
 Hydrolase      (EC 3.-.-.-)          0.135    0.425
 Lyase          (EC 4.-.-.-)          0.049    1.054
 Isomerase      (EC 5.-.-.-)          0.010    0.321
 Ligase         (EC 6.-.-.-)          0.042    0.827


 Gene Ontology category               Prob     Odds
 Signal_transducer                    0.201    0.939
 Receptor                             0.353    2.076
 Hormone                              0.002    0.365
 Structural_protein                   0.005    0.190
 Transporter                          0.024    0.219
 Ion_channel                          0.008    0.147
 Voltage-gated_ion_channel            0.002    0.085
 Cation_channel                       0.010    0.221
 Transcription                        0.036    0.283
 Transcription_regulation             0.018    0.147
 Stress_response                      0.274    3.108
 Immune_response                   => 0.381    4.486
 Growth_factor                        0.013    0.943
 Metal_ion_transport                  0.009    0.02

Reference

<references />

Retrieved from "https://i12r-studfilesrv.informatik.tu-muenchen.de/wiki/index.php?title=Sequence-based_predictions&oldid=4619"