Difference between revisions of "Task 3 (MSUD)"
m (→Discussion) |
(→Discussion) |
||
| Line 484: | Line 484: | ||
Other methods for GO term prediction are [http://kinaz.fen.bilkent.edu.tr/gopred/ GOPred], [http://www.blast2go.com/b2ghome Blast2GO] and [http://eagl.unige.ch/GOCat/ GOCat]. |
Other methods for GO term prediction are [http://kinaz.fen.bilkent.edu.tr/gopred/ GOPred], [http://www.blast2go.com/b2ghome Blast2GO] and [http://eagl.unige.ch/GOCat/ GOCat]. |
||
| + | |||
| + | |||
| + | |||
| + | From protein sequence alone, additional features can be predicted: |
||
| + | * solvent accesibility |
||
| + | * posttranslational modifications |
||
| + | * localization |
||
| + | * metal binding sites |
||
| + | * active sites |
||
| + | * disulfide bridges |
||
| + | * SNP effects |
||
Revision as of 12:23, 2 August 2013
Contents
Secondary structure
Result
Approach for predicting secondary structure with ReProf
For P10775, ReProf was run with the protein sequence fasta file and position specific scoring matrices (PSSM) derived from big_80 and SwissProt as input. The following tables show the comparison of the prediction results to the secondary structure assignment of DSSP. The f-measure is the harmonic mean of recall and precision, it gives a good indication for the quality of a classificator.
| secondary structure element | recall | precision | f-measure |
|---|---|---|---|
| H | 0.719 | 0.585 | 0.645 |
| E | 0.211 | 0.500 | 0.296 |
| L | 0.616 | 0.654 | 0.635 |
| secondary structure element | recall | precision | f-measure |
|---|---|---|---|
| H | 0.944 | 0.889 | 0.916 |
| E | 0.649 | 0.685 | 0.667 |
| L | 0.826 | 0.866 | 0.846 |
| secondary structure element | recall | precision | f-measure |
|---|---|---|---|
| H | 0.923 | 0.914 | 0.919 |
| E | 0.807 | 0.523 | 0.634 |
| L | 0.719 | 0.859 | 0.782 |
Predictions using a PSSM instead of a simple sequence have a considerably better quality. All methods predict helices better than loops and these better than beta sheets. The results of the run with the big_80 PSMM are better for E and L and only slightly worse for H than those using the SwissProt PSMM.
The percentages of correctly identified secondary structure (H, E or L) for the three methods are 61 %, 86 % and 82 %. So for the remaining sequences, the method with the best performance (usage of PSSM derived from big_80 as input for ReProf) was used.
Comparison of ReProf to PsiPred and DSSP
The following tables show the percentages of agreement for secondary structure between ReProf and PsiPred or DSSP.
P12694
| secondary structure element | ReProf vs. PsiPred | ReProf vs. DSSP |
|---|---|---|
| H | 0.804 | 0.812 |
| E | 0.400 | 0.585 |
| L | 0.876 | 0.782 |
| all | 0.849 | 0.816 |
P10775
| secondary structure element | ReProf vs. PsiPred | ReProf vs. DSSP |
|---|---|---|
| H | 0.798 | 0.889 |
| E | 0.691 | 0.649 |
| L | 0.779 | 0.828 |
| all | 0.849 | 0.855 |
Q08209
| secondary structure element | ReProf vs. PsiPred | ReProf vs. DSSP |
|---|---|---|
| H | 0.794 | 0.816 |
| E | 0.487 | 0.615 |
| L | 0.830 | 0.807 |
| all | 0.827 | 0.807 |
Q9X0E6
| secondary structure element | ReProf vs. PsiPred | ReProf vs. DSSP |
|---|---|---|
| H | 0.897 | 0.923 |
| E | 0.694 | 0.643 |
| L | 0.636 | 0.545 |
| all | 0.802 | 0.802 |
Altogether, ReProf agrees in 80-85% of the predictions with PsiPred and DSSP. In most cases the agreement for H and L is higher than for E.
Information from UniProt and PDB
A summary of interesting features for the proteins taken from UniProt and PDB:
P12694, 2BFD:
- name: 2-oxoisovalerate dehydrogenase subunit alpha, mitochondrial
- EC: 1.2.4.4
- gene: BCKDHA
- organism: Homo sapiens (Human)
- sequence length: 445 AA
- subunit structure: heterotetramer of alpha and beta chains
- subcellular location: mitochondrion matrix
- secondary structure: 42% helical, 10% beta sheet
- 3D similarity: pyruvate dehydrogenase E1
- ligands: chloride ion, glycerol, potassium ion, manganese (II) ion, (4S)-2-methyl-2,4-pentanediol, thiamin diphosphate
P10775, 2BNH:
- name: ribonuclease inhibitor
- gene: RNH1
- organism: Sus scrofa (Pig)
- sequence length: 456 AA
- subcellular location. cytoplasm
- sequence similarities: contains 15 LRR (leucine-rich) repeats
- secondary structure: alternating helix and strand, 42% helical, 12% beta sheet
Q08209, 1AUI:
- name: serine/threonine-protein phosphatase 2B catalytic subunit alpha isoform
- EC: 3.1.3.16
- gene: PPP3CA
- organism: Homo sapiens (Human)
- sequence length: 521 AA
- subunit structure: heterodimer of alpha and beta chain (human calcineurin heterodimer)
- subcellular location: nucleus
- secondary structure: 27% helical, 11% beta sheet
- ligands: calcium ion, Fe (III) ion, zinc ion
Q9X0E6, 1KR4:
- name: divalent-cation tolerance protein CutA
- gene: cutA
- organism: Thermotoga maritima (strain ATCC 43589 / MSB8 / DSM 3109 / JCM 10099)
- sequence length: 101 AA
- subunit structure: homotrimer
- subcellular location: cytoplasm
- secondary structure: great fraction of strands, 29% helical, 35% beta sheet
Discussion
In the first step, the ReProf results for P10775 were evaluated against the DSSP assignment. Here, DSSP was viewed as a standard of truth, because it assigns secondary structure based on the measured 3D structure by examining angles and H bonds between atoms.
The prediction of secondary structure is much better if a PSSM is used instead of the sequence. The reason is that a PSSM describes the requirements for each position better than the amino acid sequence, because it uses evolutionary information. So it identifies for each position alternatives for the residues in the primary sequence, that do not alter the overall structure of the protein. The difference between the usage of big_80 or SwissProt for generating the PSSM is not that obvious, but we decided to take big_80 for the remaining proteins because it showed a slightly better performance in our test with the example protein P10775.
For all proteins these ReProf results were compared to PsiPred and DSSP, to see how much the methods agree in the secondary structure assignment. The methods agree for most residues. The highest discrepancies can be observed for the prediction of beta strands, which shows that these are not as easy to predict as alpha helices or loops. A reason for this could be that beta strands are less frequent than alpha helices in most proteins, as can be seen in the informations taken from UniProt and PBD.
Other method for prediction of secondary structure are GOR, Jpred and PHDsec. Another secondary structure assignment tool is STRIDE.
Disordered protein
Result
IUPred
IUPred performs prediction of intrinsic disordered regions of proteins by the observation of pairwise energy content from proteins with known structures. The results of IUPred fall into 3 categories: long (disordered regions), short (disordered regions) and globular (regions of protein where pairwise interactions between residues are more potential).
For the prediction of disordered regions (prediction mode long and short), position-specific score is assigned to each residue of the query sequence. The score indicates the tendency that the corresponding residue belongs to a disordered region. Following are the profiles of disorder tendency generated by web-tool of IUPred:
BCKDHA
- Profiles for disorders and globular regions of BCKDHA
Profile of long range and short range disorders
Difference between scores of long range disorders and short range disorders
Prediction of globular regions
P10775
- Profiles for disorders and globular regions of P10775
Profile of long range and short range disorders
Difference between scores of long range disorders and short range disorders
Prediction of globular regions
Q9X0E6
- Profiles for disorders and globular regions of Q9X0E6
Profile of long range and short range disorders
Difference between scores of long range disorders and short range disorders
Prediction of globular regions
Q08209
- Profiles for disorders and globular regions of Q08209
Profile of long range and short range disorders
Difference between scores of long range disorders and short range disorders
Prediction of globular regions
Statistics
- Relations between profiles of long and short disorders
Metadisorder
- Profile of raw scores and predicted globular regions
Predicted disordered and globular regions of BCKDHA
Predicted disordered and globular regions of P10775
Predicted disordered and globular regions of Q9X0E6
Predicted disordered and globular regions of Q08209
DisProt
Because DisProt dose not contain all uniprot sequences, we have searched for homologous sequences in DisProt for the protein sequences.
- Disordered regions found in DisProt (yellow: whole protein, red:disordered regions, blue:ordered regions)
Homologous sequence for BCKDHA in DisProt (E-value: 0.19). Homologous region is 47-127 (in BCKDHA 351-432).
Homologous sequence for P10775 in DisProt (E-value: 1.4e-24). Homologous region is 780-985 (in P10775 248-456).
Homologous sequence for Q9X0E6 in DisProt (E-value: 0.34). Homologous region is 10-90 (in Q9X0E6 12-101).
There is an entry for Q08209 in DisProt.
Discussion
IUPred
- Generally the profiles of long and short disorders are similar because they are overall highly correlated (except for protein Q9X0E6).
- At both ends of the protein short disorder scores are much higher. This reflects the fact that residues at ends of proteins have higher spatial flexibility and are structurally more unstable.
- High short disorder scores of Q9X0E6 can be explained by its structure.
- As is shown in its X-ray structure, at the both ends of the thick representation of backbone indicates high B-values which mean the conformation of residues is either very flexible or even undefined.
- While red parts of the protein show symmetry-related crystal contacts (within 5 Å), the remaining parts have barely symmetry-related crystal contacts.
- The overall lower long disorder score may be explained by the fact that the protein still fold into a definite 3D structure, despite of local flexibility.
- Q9X0E6 also falls into different category in comparison to the other 3 proteins.
- As is described in InterPro its function is not clear but should have a role in signal transduction.
- The other 3 proteins are either subunits or inhibitor of enzymes.
MetaDisorder
- In comparison to IUPred, MetaDisorder seems to be more sensitive to input data. It finds out more short structural regions that locate in disordered regions.
- For sequence Q08209, the prediction result of MetaDisorder shows almost the same disordered regions which are annotated in DisProt.
- Generally IUPred and MetaDisorder share very similar results.
- For sequence Q9X0E6, the predicted results are most dissimilar. The annotation of homologous sequence in DisProt can not show significant information for comparison to prediction methods because the E-value of local alignment is too high.
Comparison between Prediction and Annotation
- Comparison between prediction results of IUPred, MetaDisorder and annotation in Disprot
Annotation of disordered regions in protein P50224. (Yellow: whole protein, Red: disordered regions, Blue: ordered regions)
Prediction result from IUPred for protein P50224.
Prediction result from MetaDisorder for protein P50224.
The prediction results of Human Sulfotransferase 1A3/1A4(P50224) are dissimilar to the annotations in DisProt. It seems structural features like B-factors and symmetry-related crystal contacts give important clue to intrinsically disordered regions. As is shown in PDB structure of P50224, large range of amino-acids have high B-factors.
Transmembrane helices
Result
PolyPhobius
Except BCKDHA all other proteins were predicted to have transmembrane helices. Although there is a weak signal for transmembrane region, the protein P35462 is predicted to have 6 transmembrane helices. The protein P47863 is predicted to have 6 transmembrane helices. And protein Q9YDF8 is predicted to have 7 transmembrane helices.
- Prediction results of PolyPhobius
Residue localization of BCKDHA predicted by PolyPhobius
Residue localization of P35462 predicted by PolyPhobius. 6 transmembrane helices.
Residue localization of P47863 predicted by PolyPhobius. 6 transmembrane helices.
Residue localization of Q9YDF8 predicted by PolyPhobius. 7 transmembrane helices.
MEMSAT-SVM
BCKDHA is predicted to have 1 transmembrane helix. All the other proteins are predicted to have 6 transmembrane helices.
- Predicted topology of transmembrane helices by MEMSAT-SVM
Prediction result for protein BCKDHA
Prediction result for protein P35462
Prediction result for protein P47863
Prediction result for protein Q9YDF8
Annotations in OPM and PDBTM
- BCKDHA: no annotation was found in OPM and PDBTM.
- P35462(3PBL):
- Orientation of transmembrane helices in OPM and PDBTM
Structural annotation of P35462 in OPM. There are total 7 transmembrane helices.
Structural annotation of P35462 in PDBTM. There are total 7 transmembrane helices.
- P47863(2D57): The PDB structure is a homo tetramer. Protein P47863 is one of the 4 identical chains.
- Orientation of transmembrane helices in OPM and PDBTM
Structural annotation of P47863 in OPM. There are total 32 transmembrane helices. Each chain has 8 transmembrane helices.
Structural annotation of P47863 in PDBTM. There are total 32 transmembrane helices. Each chain has 6 transmembrane helices.
- Q9YDF8(2KYH):
- Orientation of transmembrane helices in OPM and PDBTM
Structural annotation of Q9YDF8 in OPM. There are total 5 transmembrane helices.
Structural annotation of Q9YDF8 in PDBTM. There are total 4 transmembrane helices.
Discussion
- Comparison between OPM and PDBTM:
- PDBTM seems to have more narrow transmembrane region than the OPM.
- For same structure, PDBTM tends to assign less transmembrane helices in comparison to OPM.
- As we already know, BCKDHA is a intra-mitochondrial protein. The prediction result of MEMSAT-SVM is wrong.
- Generally prediction results of PolyPhobius and MEMSAT-SVM are similar to annotations in OPM and PDBTM.
Signal peptides
Result
SignalP predictions
The following diagrams show the C- (cleavage site), S- (signal peptide) and Y- (combined cleavage site) scores for the three proteins according to SignalP version 4.1. The combined cleavage site score combines the cleavage score with the slope of the signal peptide score to optimize the recognition of cleavage sites.
For P02768 and P11279 signal peptides are predicted, with the clevage site between position 18 and 19 for P02768 and between 28 and 29 for P11279. For P47863, no signal peptide is predicted.
SignalP version 3.0 came to the same result for P02768 and P11279. However for P47863 it predicts a signal peptide with cleavage site between positions 54 and 55 (neural network) or 56 and 57 (hidden markov model), although only with the probability 0.723 compared to near 1 for the other two proteins.
Known signal peptides
On the Signal Peptide Website there are entries for P02768 and P11279 but not for P47863:
| Accession Number | Entry Name | Protein Name | Organism | Length | Status | Signal Sequence |
|---|---|---|---|---|---|---|
| P02768 | ALBU_HUMAN | Serum albumin | Homo sapiens | 18 | confirmed | MKWVTFISLLFLFSSAYS |
| P11279 | LAMP1_HUMAN | Lysosome-associated membrane glycoprotein 1 | Homo sapiens | 28 | confirmed | MAAPGSARRPLLLLLLLLLLGLMHCASA |
Discussion
The predictions of the newest version of SignalP agree with the confirmed signal peptides. The older version predicted a signal peptide in P47863, where no one is. According to UniProt, P47863 has transmembrane helices, so these might be mistaken for a signal peptide by the old version because they resemble each other.
Other methods for signal peptide prediction are Phobius, PrediSi and Signal-3L.
GO terms
Result
GOPET
| GOid | Aspect | Confidence | GO term |
|---|---|---|---|
| GO:0003824 | F | 97% | catalytic activity |
| GO:0016491 | F | 96% | oxidoreductase activity |
| GO:0016624 | F | 95% | oxidoreductase activity acting on the aldehyde or oxo group of donors disulfide as acceptor |
| GO:0003863 | F | 90% | 3-methyl-2-oxobutanoate dehydrogenase 2-methylpropanoyl-transferring activity |
| GO:0004739 | F | 89% | pyruvate dehydrogenase acetyl-transferring activity |
| GO:0004738 | F | 78% | pyruvate dehydrogenase activity |
| GO:0003826 | F | 77% | alpha-ketoacid dehydrogenase activity |
| GO:0047101 | F | 75% | 2-oxoisovalerate dehydrogenase acylating activity |
| GO:0008677 | F | 65% | 2-dehydropantoate 2-reductase activity |
| GO:0019152 | F | 63% | acetoin dehydrogenase activity |
| GO:0030955 | F | 63% | potassium ion binding |
| GO:0016616 | F | 62% | oxidoreductase activity acting on the CH-OH group of donors NAD or NADP as acceptor |
| GO:0046872 | F | 62% | metal ion binding |
Most terms concerning the catalytic activity (GO:0003824, GO:0016491, GO:0016624, GO:0003863, GO:0003826, GO:0047101) are consitent with the knowledge about the enzyme activity of the 2-oxoisovalerate dehydrogenase (see biochemical description of MSUD). Also the terms about metal binding (GO:0030955, GO:0046872) correspond to the characterization in PDB.
ProtFun
############## ProtFun 2.2 predictions ############## >gi_11386135 # Functional category Prob Odds Amino_acid_biosynthesis 0.187 8.520 Biosynthesis_of_cofactors 0.246 3.413 Cell_envelope 0.035 0.581 Cellular_processes 0.041 0.560 Central_intermediary_metabolism => 0.321 5.096 Energy_metabolism 0.208 2.310 Fatty_acid_metabolism 0.023 1.738 Purines_and_pyrimidines 0.257 1.059 Regulatory_functions 0.031 0.194 Replication_and_transcription 0.170 0.636 Translation 0.047 1.078 Transport_and_binding 0.029 0.071 # Enzyme/nonenzyme Prob Odds Enzyme => 0.769 2.683 Nonenzyme 0.231 0.324 # Enzyme class Prob Odds Oxidoreductase (EC 1.-.-.-) 0.178 0.857 Transferase (EC 2.-.-.-) 0.238 0.690 Hydrolase (EC 3.-.-.-) 0.190 0.601 Lyase (EC 4.-.-.-) 0.076 1.614 Isomerase (EC 5.-.-.-) 0.010 0.321 Ligase (EC 6.-.-.-) => 0.085 1.673 # Gene Ontology category Prob Odds Signal_transducer 0.098 0.458 Receptor 0.006 0.038 Hormone 0.001 0.206 Structural_protein 0.005 0.170 Transporter 0.025 0.226 Ion_channel 0.009 0.163 Voltage-gated_ion_channel 0.004 0.170 Cation_channel 0.010 0.215 Transcription 0.060 0.470 Transcription_regulation 0.053 0.427 Stress_response 0.010 0.110 Immune_response 0.012 0.136 Growth_factor 0.009 0.609 Metal_ion_transport 0.012 0.025 //
The protein is predicted to have a function in the central intermediary metabolism, be an enzyme and belong to the enzyme class EC 6 (ligase). The last aspect is not correct, since it is a oxidoreductase (compare Information from UniProt and PDB).
Pfam
Pfam sequence search gives one significant Pfam-A match:
- E1_dh (dehydrogenase E1 component, PF00676)
- use thiamine pyrophosphate as cofactor
- includes pyruvate dehydrogenase, 2-oxoglutarate dehydrogenase and 2-oxoisovalerate dehydrogenase
- members of multienzyme complex
- interactions: E1_dh, Transketolase_C (transketolase, C-terminal domain), Transket_pyr (transketolase, pyrimidine binding domain)
- 9023 sequences
- belongs to clan THDP-binding (CL0254): thiamin diphosphate-binding superfamily
Discussion
The GOPET results give a good overview about the catalytic and binding activities of our protein, of which most are consitstent with the current knowledge. Whether the enzyme can really accept the other substrates pyruvate, 2-dehydropantoate and acetoin, as the prediction suggests, is not clear. It is possible that these activities were only predicted because of the high sequence identity to pyruvate dehydrogenase and other dehydrogenases. The fact that the confidences for these predictions are not as high as for some of the others, argues for this interpretation.
That ProtFun predicted the false enzyme class for our protein shows that this prediction is not always easy. In support of the method it has to be stated, that the probability and odds values for the different enzyme classes are close to each other. Also these values for the "right" enzyme class (EC 1) are some of the higher ones.
Pfam helps to find related proteins that are clustered into families that have common domains. Families are grouped together in clans, so one can also find out more about the distant relationship between proteins. Pfam found a family with different dehydrogenases from diverse organisms that are homolog to our protein. They have in common to be the first part of a large enzyme complex. So the concept of oxidative decarboxylation has adapted to different substrates and different organisms during evolution, but still uses the same basic principle.
Other methods for GO term prediction are GOPred, Blast2GO and GOCat.
From protein sequence alone, additional features can be predicted:
- solvent accesibility
- posttranslational modifications
- localization
- metal binding sites
- active sites
- disulfide bridges
- SNP effects
