Fabry:Sequence-based analyses

From Bioinformatikpedia

Fabry Disease » Sequence-based analyses



The following analyses were performed on the basis of the α-Galactosidase A sequence. Please consult the journal for the commands used to generate the results.

Secondary structure

In the following, the sequence-based secondary structure predictions programs ReProf and PsiPred are applied on four different proteins and its predictions are compared to each other and the DSSP reference. DSSP is a good way to measure the prediction performance since it is not sequence-based but operates on the three-dimensional orientation of the protein atoms as they are provided by a PDB file.

ReProf does not seem to post-process its results to remove artefacts. In contrary to the PsiPred results, there are several helix breaks by a single Coil or Beta-Sheet residue.

For the comparisons, there were the following adjustment madeDSSP results had to be adjusted in the following since some of the PDB files represent slightly longer sequences than the fasta sequence that were used as input for the sequence-based prediction methods. So some residues at the termini, that were not part of the FASTA sequence, were removed and some gaps at the beginning and the end were added

Alpha-galactosidase A (P06280)

Alpha-galactosidase A
Alpha GAL A.png
Organism Homo sapiens (Human)
Subcellular location Lysosome
Sequence length 429 AA
SCOP Alpha and beta proteins (a/b)
All beta proteins
UniProt-AC P06280
PDB-ID 1R46

For the comparison, there were some gaps added to the DSSP secondary structure, because the DSSP output did not include all the residues of the sequence.

Prediction method Q3 SOV
PsiPred 79.02% 63.00%
ReProf 60.84% 53.13%

<xr id="fig:P06280_comaprison"/> shows the differences between the ReProf and PsiPred predictions and the DSSP reference.

<figure id="fig:P06280_comaprison">

Comparison of the predicted secondary structures of ReProf and PsiPred for P06280 and the DSSP reference.

</figure>


Ribonuclease inhibitor (P10775)

Ribonuclease inhibitor
Fabry 2BNH.jpg
Organism Sus scrofa (Pig)
Subcellular location Cytoplasm
Sequence length 456 AA
SCOP Alpha and beta proteins (a/b)
UniProt-AC P10775
PDB-ID 2BNH
Prediction method Q3 SOV
PsiPred 91.89% 95.47%
ReProf 60.96% 66.03%

<xr id="fig:P10775_comaprison"/> shows the differences between the ReProf and PsiPred predictions and the DSSP reference.

<figure id="fig:P10775_comaprison">

Comparison of the predicted secondary structures of ReProf and PsiPred for P10775 and the DSSP reference.

</figure>


Serine/threonine-protein phosphatase 2B catalytic subunit alpha isoform (Q08209)

Serine/threonine-protein phosphatase 2B
catalytic subunit alpha isoform
Fabry 1AUI.jpg
Organism Homo sapiens (Human)
Subcellular location Nucleus
Sequence length 521 AA
SCOP Alpha and beta proteins (a+b)
All alpha proteins
UniProt-AC Q08209
PDB-ID 1AUI

For the comparison, there were some gaps added to the DSSP secondary structure, because the DSSP output did not include all the residues of the sequence.

Prediction method Q3 SOV
PsiPred 74.66% 54.32%
ReProf 57.97% 30.66%

<xr id="fig:Q08209_comaprison"/> shows the differences between the ReProf and PsiPred predictions and the DSSP reference.

<figure id="fig:Q08209_comaprison">

Comparison of the predicted secondary structures of ReProf and PsiPred for Q08209 and the DSSP reference.

</figure>


Divalent-cation tolerance protein CutA (Q9X0E6)

Divalent-cation tolerance protein CutA
Fabry 1KR4.jpg
Organism Thermotoga maritima
(Bacteria)
Subcellular location Cytoplasm
Sequence length 101 AA
SCOP Alpha and beta proteins (a+b)
UniProt-AC Q9X0E6
PDB-ID 1KR4

The DSSP output needed to be adjusted since the sequence represented by the PDB file was slightly longer. For this reason, the first seven and the last two residues were removed to match the sequence from the Q9X0E6 Uniprot entry.

Prediction method Q3 SOV
PsiPred 85.15% 82.61%
ReProf 57.43% 60.37%

<xr id="fig:Q9X0E6_comaprison"/> shows the differences between the ReProf and PsiPred predictions and the DSSP reference.

<figure id="fig:Q9X0E6_comaprison">

Comparison of the predicted secondary structures of ReProf and PsiPred for Q9X0E6 and the DSSP reference.

</figure>


Disorder

For the disorder predictions, IUPred was used. IUPred supports three different types of predictions: globular domains (glob), long disorder (long) and short disorder (short). Since it predicts for three out of the four protein we analyzed only one globular domain which ranges from the first to the last residues, we do not consider this type of prediction in the following sections. The exception is Q08209 where the single predicted globular domain ranges from 5 - 446 with a total sequence length of 521 amino acids.

Alpha-galactosidase A (P06280)

<figure id="fig:P06280_long_disorder">

Long disorder likelihoods per residue as predicted by IUPred. Residues above the red line (0.5) are assumed as disordered.

</figure> <figure id="fig:P06280_short_disorder">

Short disorder likelihoods per residue as predicted by IUPred. Residues above the red line (0.5) are assumed as disordered.

</figure>


For α-Galactosidase A, IUPred predicts two long disordered residues as shown in <xr id="fig:P06280_long_disorder"/>:

107 Q     0.5055
108 A     0.5533

<xr id="fig:P06280_short_disorder"/> shows the short disorder likelihoods. There are 15 short disordered residues in AGAL. At the N-terminus (1 - 4), the C-terminus (425 - 429) and around the 100th residue (102, 106 - 110).

There is no corresponding entry for the Alpha-galactosidase A protein in Disprot and there was also no significant hit found using the provided Psi-Blast search:

                                                                  Score    E
Sequences producing significant alignments:                      (bits) Value
DP00151                                                             25   1.6
DP00110                                                             25   2.8
DP00121                                                             24   3.4

Ribonuclease inhibitor (P10775)

<figure id="fig:P10775_long_disorder">

Long disorder likelihoods per residue as predicted by IUPred. Residues above the red line (0.5) are assumed as disordered.

</figure> <figure id="fig:P10775_short_disorder">

Short disorder likelihoods per residue as predicted by IUPred. Residues above the red line (0.5) are assumed as disordered.

</figure>


IUPred predicts no long disorder residues for the ribonuclease inhibitor (see <xr id="fig:P10775_long_disorder"/>), but there are ten short disordered residues at the N-terminus (1-5) and the C-terminus (452 - 456) as plotted in <xr id="fig:P10775_short_disorder"/>.

There is no corresponding entry for the protein in Disprot, but there is one single significant hit DP00554:

                                                                 Score    E
Sequences producing significant alignments:                      (bits) Value
DP00554                                                            123   5e-30
DP00489                                                             32   0.022
DP00191                                                             27   0.79
DP00252                                                             25   2.4
DP00102                                                             25   2.4
DP00474                                                             24   4.0
DP00087                                                             24   4.6
DP00072                                                             23   8.8

<figure id="fig:P10775_disprot">

Scheme of the disordered region of Disprot entry DP00554

</figure>

According to the Disprot entry, there is a 20aa-long disordered region from residue 31 to 50 (shown in <xr id="fig:P10775_disprot"/>), which does not comply with the IUPred predictions located at the protein termini. However, ribonuclease inhibitor (P10775) is from pig and the Disprot entry describes Q9C000-1 from human.


Serine/threonine-protein phosphatase 2B catalytic subunit alpha isoform (Q08209)

<figure id="fig:Q08209_long_disorder">

Long disorder likelihoods per residue as predicted by IUPred. Residues above the red line (0.5) are assumed as disordered.

</figure> <figure id="fig:Q08209_short_disorder">

Short disorder likelihoods per residue as predicted by IUPred. Residues above the red line (0.5) are assumed as disordered.

</figure>


According to IUPred, there are 81 long disordered residues and 79 short disordered residues (see <xr id="fig:Q08209_short_disorder"/>) in Serine/threonine-protein phosphatase 2B catalytic subunit alpha isoform. As it can be seen in <xr id="fig:Q08209_long_disorder"/>, the long disordered residues are located at the N-terminus (1 - 8) and from residue 383 upward (383 - 384; 424 - 425; 432, 434 - 439; 443, 445, 448 - 449; 455, 458; 463 - 466), with a large region at the C-terminus from 469 - 521.

<figure id="fig:Q08209_disprot">

Scheme of the disordered region of Disprot entry DP00554

</figure>

The predicted short and long disordered regions share a large amount of residues, especially at the N- (1 - 8 long vs. 1 - 11 short) and C-termini (both share 463 - 466 and 469 - 521). There are some more short disordered residues near the N-terminus than long disordered ones (13, 18 - 19, 24, 32 - 35) and less near the C-terminus (only 434, 437 and 460 in addition to the already above described common ones).

For this protein, there is an associated Disprot entry at DP00092 which graphical scheme is shown in <xr id="fig:Q08209_disprot"/>. Its annotations mostly accord with the IUPred predictions, in particular at the termini.


Divalent-cation tolerance protein CutA (Q9X0E6)

<figure id="fig:Q9X0E6_long_disorder">

Long disorder likelihoods per residue as predicted by IUPred. Residues above the red line (0.5) are assumed as disordered.

</figure> <figure id="fig:Q9X0E6_short_disorder">

Short disorder likelihoods per residue as predicted by IUPred. Residues above the red line (0.5) are assumed as disordered.

</figure>


There are no long disordered residues according to IUPred in Divalent-cation tolerance protein CutA (see <xr id="fig:Q9X0E6_long_disorder"/>), but six short disordered residues (<xr id="fig:Q9X0E6_long_disorder"/>), again at the termini.

  1 M     0.6604
  2 I     0.6293
  3 L     0.5900
  4 V     0.5412
100 V     0.5549
101 L     0.5941

There is no corresponding entry for this protein in Disprot and there was also no significant hit found using the provided Psi-Blast search:

                                                                 Score    E
Sequences producing significant alignments:                      (bits) Value
DP00101                                                             22   2.7
DP00175                                                             21   4.3
DP00046                                                             21   4.3
DP00012                                                             20   6.9
DP00055                                                             20   7.1

Transmembrane helices

α-Galactosidase A (P06280)

<figure id="fig:1r46_TM">

Human α-Galactosidase A, plot of posterior probabilities of features (source)

</figure> <figure id="fig:1r46_TMHMM">

Transmembrane helix prediction with TMHMM for human α-Galactosidase A

</figure>

organism: Homo sapiens (Human)
pdb-id: 1r46

Since the whole sequence is labeled as "NON CYTOPLASMIC", the prediction of Polyphobius is that it contains no membrane helices (see Polyphobius output file). This result is consistent with both databases, OPM and PDBTM and the TMHMM-2.0 prediction (see <xr id="fig:1r46_TMHMM" />).
In <xr id="fig:1r46_TM"/> the posterior probabilities of cytoplasmic(green), non cytoplasmic(blue), TM helix(grey area) and signal peptide(red) are shown. The grey area indicates weak evidence for a transmembrane helix which where not predicted. The first small peak can be observed in both pictures. Considering the shown probabilities and our background knowledge, the prediction seems to be true. Only the end of the signal peptide at residue 29 is, according to our knowledge, too early and should be shifted to position 31.


D(3) dopamine receptor (P35462)

<figtable id="tab:3pbl_TM_PostProb"> Posterior probability plots Polyphobius and TMHMM

Human D(3) dopamine receptor, plot of posterior probabilities of features predicted by Polyphobius
Human D(3) dopamine receptor, plot of posterior probabilities of features predicted by TMHMM

</figtable>

organism: Homo sapiens (Human)
pdb-id: 3pbl (only available structure)

Polyphobius predicts 7 transmembrane regions. Comparing this result to the pictures in <xr id="tab:3pbl_TM"/>, we can again see a consistent result with the two databases. The only difference between the models seems to be the number of residues on the cytosolic side of the membrane (green in the right picture) of this hydrolase/hydrolase inhibitor. Inquiring the PDBTM site, this area is marked as residues 1002-1161, which do not even belong to the 400 aa long protein. This might result from the experiment, the structure was derived from. It says, that the protein was enginered (see source1 and source2)
Besides from that, a difference in predicted and observed starting and end points could be revealed (see <xr id="tab:3pbl_start_end"/>), mainly those of PDBTM, which leads to a great difference in its mean length compared to the other methods (see <xr id="tab:3pbl_TM"/>, picture 3). Here, PDBTM throughout assigns shorter helices, since its helices always start later and end earlier, which can also be seen in <xr id="fig:bitshift" />. The boundaries Polyphobius predicted may be seen in the Polyphobius output file.
The posterior probability of the first 6 predicted helices is very similar, as one can see in <xr id="tab:3pbl_TM_PostProb"/>. The only difference is in the last TM, where the probability assigned by TMHMM is nearly 90%, whereas Polyphobius only assigns about 55%.

<figtable id="tab:3pbl_start_end"> Start and end points of each transmembrane helix predicted

TMH number 1 2 3 4 5 6 7
Method Length Start End Length Start End Length Start End Length Start End Length Start End Length Start End Length Start End
Polyphobius 26 30 55 23 66 88 22 105 126 21 150 170 25 188 212 24 329 352 20 367 386
OPM 19 34 52 25 67 91 26 101 126 21 150 170 23 187 209 22 330 251 24 363 382
PDBTM 18 35 52 17 68 84 15 109 123 15 152 166 16 191 206 14 334 347 15 368 382
Uniprot 23 33 55 23 66 88 22 105 126 21 150 170 25 188 212 22 330 351 22 367 388
TMHMM 23 32 54 23 67 89 23 104 126 23 150 172 23 192 214 23 331 353 23 368 390

</figtable>


<figtable id="tab:3pbl_TM"> Visualizations for the human D(3) dopamine receptor as a transmembrane protein. Note: Cytoplasmic side of the membrane is different in the pictures

Human D(3) dopamine receptor as depicted in OPM database (Cytoplasmic side on the bottom) (source)
Human D(3) dopamine receptor as shown in PDBTM database (Cytoplasmic side above) (source)
Human D(3) dopamine receptor, comparison of length distribution of different prediction methods

</figtable>


Voltage-gated potassium channel (Q9YDF8)

<figtable id="tab:1orq_TM_PostProb"> Posterior probability plots Polyphobius and TMHMM for Q9YDF8

Voltage-gated potassium channel, plot of posterior probabilities of features predicted by Polyphobius
Voltage-gated potassium channel, plot of posterior probabilities of features predicted by TMHMM

</figtable>

organism: Aeropyrum pernix (strain ATCC 700893 / DSM 11879 / JCM 9820 / NBRC 100138 / K1)
pdb-id: 1orq (smaller resolution than 2AOL, more residues determined)

The voltage-gated potassium channel is a tetrameric protein, that causes a voltage-dependent potassium ion permeability of the membrane. It may occur in an opened or closed conformation. This state can be changed by the voltage difference across the membrane. (source) To fulfill this function, the protein needs the transmembrane helices. Of these, 7 are predicted according to the Polyphobius output file, but referring to OPM, only three helices should be predicted, according to PDBTM 4 helices are considered true. Looking carefully at the probability plot of Polyphobius in <xr id="tab:1orq_TM_PostProb"/>, one may see, that there actually are only four peaks of transmembrane helix posterior probability, but three of the less higher peaks are considered true as well.
Only 2 of all 7 helices are common among all three methods (see <xr id="tab:1orq_TM" />, picture 3), the third and fourth transmembrane helices are even only predicted by Polyphobius, Uniprot and TMHMM and only Polyphobius and Uniprot, respectively. If not predicted or shown by one of the methods, the helix is considered as membrane loop.
Here OPM has the smallest mean length of transmembrane helices (see <xr id="tab:1orq_TM" />, picture 3). Start and end points of the helices seem to be differing more than in the first described picture (see <xr id="tab:1orq_start_end" />, picture 3), although Polyphobius, Uniprot and TMHMM are quite similar. Personally, considering the features predicted by Polyphobius and TMHMM (see <xr id="tab:1orq_TM_PostProb" />), I would trust the PDBTM version the most in this case. But since in this case, the advantage of Polyphobius to include homologue data is not given, because the BLAST search did not find any homologue sequences, it might on the other hand be a false prediction. This might, to a certain degree, explain the rather diverse result.

<figtable id="tab:1orq_start_end"> Start and end points of each transmembrane helix predicted

TMH number 1 2 3 4 5 6 7
Method Length Start End Length Start End Length Start End Length Start End Length Start End Length Start End Length Start End
Polyphobius 19 42 60 21 68 88 22 108 129 21 137 157 22 163 184 18 196 213 21 224 244
OPM 0 - - 0 - - 0 - - 0 - - 20 153 172 13 183 195 19 207 225
PDBTM 32 21 52 24 57 80 0 - - 0 - - 21 151 171 28 209 236 0 - -
Uniprot 25 39 63 25 68 92 17 109 125 21 137 157 25 160 184 13 196 208 32 222 253
TMHMM 23 39 61 20 68 87 23 107 129 0 - - 23 162 184 20 199 218 20 225 244

</figtable>

<figtable id="tab:1orq_TM"> Visualizations for the Voltage-gated potassium channel as a transmembrane protein. Note: Cytoplasmic side of the membrane is different in the pictures

Voltage-gated potassium channel as depicted in OPM database (Cytoplasmic side on the bottom) (source)
Voltage-gated potassium channel as shown in PDBTM database (Cytoplasmic side above) (source)
Voltage-gated potassium channel, comparison of length distribution of different prediction methods

</figtable>


Aquaporin-4 (P47863)

<figtable id="tab:2d57_TM_PostProb"> Posterior probability plots Polyphobius and TMHMM for Q9YDF8

Aquaporin-4, plot of posterior probabilities of features predicted by Polyphobius
Aquaporin-4, plot of posterior probabilities of features predicted by TMHMM

</figtable>

organism: Rattus norvegicus (Rat)
pdb-id: 2d57 (non-mutant)

Another homotetramer, the water-specific channel Aquaporin-4, has been analyzed. In the first two pictures in <xr id="tab:2d57_TM"/>, the protein is depicted in its tetrameric structure, while in this exercise we only examine one of the four chains. Here Polyphobius again predicted the number of transmembrane helices correctly according to PDBTM and all the other ressources(6 - see Polyphobius output file), while OPM alone shows 8 TMH structures. Taking a closer look at the results, we see that the 2 "missing" TMHs are interpreted as membrane loop in PDBTM (depicted in orange in <xr id="tab:2d57_TM"/>, picture 2) and Polyphobius, which might be due to the short length of the helical structure (10 residues each - see <xr id="tab:2d57_TM"/>). Considering the posterior probability, I would rather prefer one of the 6 helical models.
The 6 shared helices' start and end points are again quite conform among Polyphobius and OPM, PDBTM, Uniprot and TMHMM. The mean length is again around 23, except for the two databases which have a mean length of around 18.

<figtable id="tab:2d57_start_end"> Start and end points of each transmembrane helix predicted

TMH number 1 2 3 4 5 6 7 8
Method Length Start End Length Start End Length Start End Length Start End Length Start End Length Start End Length Start End Length Start End
Polyphobius 25 34 58 22 70 91 0 - - 22 115 136 22 156 177 21 188 208 0 - - 22 231 252
OPM 23 34 56 19 70 88 10 98 107 25 112 136 23 156 178 15 189 203 10 214 223 22 231 252
PDBTM 17 39 55 18 72 89 0 - - 18 116 133 20 158 177 18 188 205 0 - - 18 231 248
Uniprot 21 37 57 21 65 85 0 - - 21 116 136 21 156 176 21 185 205 0 - - 21 232 252
TMHMM 23 33 55 23 70 92 0 - - 23 112 134 23 154 176 23 189 211 0 - - 23 231 253

</figtable>

<figtable id="tab:2d57_TM"> Visualizations for the Aquaporin-4 as a transmembrane protein. Note: Cytoplasmic side of the membrane is different in the pictures

Rat's Aquaporin-4 as depicted in OPM database (Cytoplasmic side on the bottom) (source)
Rat's Aquaporin-4 as shown in PDBTM database (Cytoplasmic side above)(source)
Aquaporin-4, comparison of length distribution of different prediction methods

</figtable>


Comparison Polyphobius, OPM, PDBTM

<figure id="fig:bitshift">

Average start and end points of transmembrane helices, Polyphobius compared to OPM, PDBTM, Uniprot and TMHMM

</figure>

In <xr id="fig:bitshift" /> the predicted transmembrane helices of Polyphobius have been used as basis and the start and end point of each TMH in the models of OPM, PDBTM, Uniprot and TMHMM (see <xr id="tab:3pbl_start_end" />, <xr id="tab:1orq_start_end" /> and <xr id="tab:2d57_start_end" />) have been compared to it. The average of each model has been calculated for each helix Polyphobius and OPM, respectively each of the other models, shared. It becomes obvious, that the transmembrane helices of OPM tend to be shifted to the left, compared to the Polyphobius helices, while the PDBTM helices usually are nested inside the Polyphobius membranes due to later starting and earlier end points. The great strength of Polyphobius becomes apparent, when examining the result for 1orq, where no homologue data could be found. (see Voltage-gated potassium channel ). Here all models are very different from Polyphobius' prediction, except for the Uniprot prediction. Taking a closer look, we found out, that the Uniprot prediction among few other methods relies on the Phobius and TMHMM algorithms (see)
Since on average, the hydrophobic belt in a membrane is about 3.5 to 5nm wide and a helix needs approximately 15 to 20 amino acids to span this width, mean lengths of 18 to 23 aas seem reasonable. Nonetheless, especially TMHMM, which seems to predict helices of length 23 in almost all cases, and also Polyphobius and Uniprot tend to have longer helices than that. This fact might be either due to the type of membrane the proteins are located at, or a bias in the algorithms.
What also stood out, was the fact that the cutoff for still counting the structure as an helical structure in the OPM database is really low, considering that an α-helix on average needs 3.6 amino-acid residues per turn. Contrariwise OPM predicts the least number of transmembrane helices in the Voltage-gated potassium channel protein, which cannot be due to the length of the predicted structure.
Summing up, the method to prefer depends on what demand you have in your result. If you want to make sure your predicted helices are transmembrane helices for sure, you want to choose PDBTM's models, since in our example, they did never assign the label TMH to a structure as the only method. But then, if you want to cover all possible structures, you should choose Polyphobius, since there has not been any transmembrane helix it did not at all predict, considering the two databases.

Signal peptides

Prediction of the presence and location of signal peptide cleavage sites in amino acid sequences.

α-Galactosidase A (P06280)

<figure id="fig:1r46_SP">

Human α-Galactosidase A, plot of scores for signal peptide cleavage site (source)

</figure>

organism: Homo sapiens (Human)
pdb-id: 1r46
Signal-peptide: 1-31
Signal-peptide sequence: MQLRNPELHLGCALALRFLALVSWDIPGARA

As already described in section Transmembrane helices the human α-Galactosidase A does not have transmembrane helical structures and from our background knowledge we know that the protein has a 31 residues long signal peptide at the N-terminus. The protein's pdb structure is available only without signal peptide, since those residues are cleaved off. Therefore no picture is provided here.
In <xr id="fig:1r46_SP"/> the signal peptide prediction of SignalP 4.0 is shown. The cleaved site is clearly indicated after residue 31, since the S-score (green) drops and the C- (red) and Y-score (blue) not only have a peak at this position, but attain their maximum at this point (see SignalP output file )



Serum Albumin (P02768)

organism: Homo sapiens (Human)
pdb-id: 1ao6
Signal-peptide: 1-18
Signal-peptide sequence: MKWVTFISLLFLFSSAYS

For this protein an 18 residues long signal peptide is predicted (see SignalP output file) and also reported in various sources (e.g. peptide website, Uniprot). The signal peptide of each chain of the homodimeric protein is shown in <xr id="tab:1ao6_SP"/> in the last picture.
According to the Polyphobius prediction, no transmembrane helix is present, but the signal peptide is also predicted. Those findings underlines the truth of the predicted signal peptide.
The rather high hydrophobicity (see <xr id="tab:1ao6_SP"/>, Figure 1) indicates an export of the protein to another cellular compartment. Examining the Uniprot website, we found that Serum Albumin is secreted into the cell surroundings, for this a membrane needs to be passed and the high hydrophobicity is needed.

<figtable id="tab:1ao6_SP"> Visualizations for the human Serum Albumin protein's signal peptide prediction.

Human Serum Albumin hydropathy plot (source)
Human Serum Albumin plot of scores for signal peptide cleavage site (source)
Human Serum Albumin, signal peptide of both chains depicted in red

</figtable>



Lysosome-associated membrane glycoprotein 1 (P11279)

organism: Homo sapiens (Human)
pdb-id: NA
Signal-peptide: 1-28
Signal-peptide sequence: MAAPGSARRPLLLLLLLLLLGLMHCASA

<figtable id="tab:P11279_SP"> Visualizations for the human Lysosome-associated membrane glycoprotein's signal peptide prediction.

Human Lysosome-associated membrane glycoprotein 1 hydropathy plot (source)
Human Lysosome-associated membrane glycoprotein 1 plot of scores for signal peptide cleavage site (source)

</figtable>

As the name implies, LAMP1 has a transmembrane helix (see Polyphobius prediction) and is associated to the plasma membrane, where it presents carbohydrate ligands to selectins. Moreover, the protein shuttles between lysosomes, endosomes, and the plasma membrane (source) and thus needs its signal peptide (residues 1-28), which has been predicted by Polyphobius and SignalP (see output). In <xr id="tab:P11279_SP"/> the hydropathy plot and a plot of the different scores used for prediction by SignalP are shown. Again, they underline the predictions of both programs.
Unfortunately, no pdb structure is available.

Aquaporin-4 (P47863)

<figure id="fig:2d57_SP">

Aquaporin-4, plot of scores for signal peptide cleavage site (source)

</figure>

organism: Rattus norvegicus (Rat)
pdb-id: 2d57
Signal-peptide: none
Signal-peptide sequence: ---

In <xr id="fig:2d57_SP"/> only a small peak of the C-score at position 41 can be observed, which by far does not reach the threshold of 0.5 for being predicted a cleavage site. All other scores are constantly low, thus no signal peptide cleavage site is predicted. As already mentioned in section Transmembrane helices, Polyphobius correctly predicts no signal peptide, but 6 transmembrane helices. This predicted (and observed) structure is in conformity with the function of a water-specific channel the porin fulfills.

GO terms

QuickGO

<figtable id="tab:QuickGO"> Results of the QuickGO search

Code Name
GO:0052692 raffinose alpha-galactosidase activity
GO:0051001 negative regulation of nitric-oxide synthase activity
GO:0046479 glycosphingolipid catabolic process
GO:0046477 glycosylceramide catabolic process
GO:0045019 negative regulation of nitric oxide biosynthetic process
GO:0044281 small molecule metabolic process
GO:0043202 lysosomal lumen
GO:0043169 cation binding
GO:0042803 protein homodimerization activity
GO:0016936 galactoside binding
GO:0016798 hydrolase activity, acting on glycosyl bonds
GO:0016787 hydrolase activity
GO:0016139 glycoside catabolic process
GO:0009311 oligosaccharide metabolic process
GO:0008152 metabolic process
GO:0006687 glycosphingolipid metabolic process
GO:0006665 sphingolipid metabolic process
GO:0005975 carbohydrate metabolic process
GO:0005794 Golgi apparatus
GO:0005764 lysosome
GO:0005737 cytoplasm
GO:0005625 soluble fraction
GO:0005576 extracellular region
GO:0005515 protein binding
GO:0005102 receptor binding
GO:0004557 alpha-galactosidase activity
GO:0004553 hydrolase activity, hydrolyzing O-glycosyl compounds
GO:0003824 catalytic activity

</figtable>

Since we used QuickGO in Task2 to download the GO terms we decided to refine our GO analysis now. The QuickGO search revealed 28 distinct GO terms (see <xr id="tab:QuickGO"/>). A lot of these are about hydrolase activity and metabolic processes (especially glycosphingolipid biosynthesis), as well as the location of the protein (lysosome) and its enzymatic function. The rather large number of resulting GO terms, may be explained by the listing of all possible terms, especially the quite general ones.


GOPET

<figtable id="tab:GOPET"> Results of the GOPET search

Result for GOPET search
GOid Aspect Confidence GO term
GO:0016798 Molecular Function Ontology (F) 98% hydrolase activity acting on glycosyl bonds
GO:0004553 Molecular Function Ontology (F) 98% hydrolase activity hydrolyzing O-glycosyl compounds
GO:0016787 Molecular Function Ontology (F) 97% hydrolase activity
GO:0004557 Molecular Function Ontology (F) 96% alpha-galactosidase activity
GO:0008456 Molecular Function Ontology (F) 89% alpha-N-acetylgalactosaminidase activity

</figtable>

Searching the GOPET annotation tool with the AGAL_HUMAN sequence revealed 5 GOIds, which are displayed in <xr id="tab:GOPET"/>. Even maximizing the "maximum number of GO prediction to be displayed per sequence" and minimizing the "confidence threshold for prediction" did not result in any more sequences.
On a first glance, since we already know the name and function of the protein, it is a bit surprising, that alpha-galactosidase activity is only the third entry with 96% confidence. In our already carried out information gathering we learned that α-galactosidase A is a hydrolase thus the first three entries were not surprising. Considering that our enzyme mainly is a glycosidase, the both entries on top of the list make perfekt sense.
Again a bit surprising was the last entry. α-N-Acetylgalactosaminidase is actually used for enzyme replacement therapy, which we mention on our main page. The structure of both enzymes is similar to each other, but this still does not explain the association of this GO term to the AGAL protein. Also, the QuickGO search did not bring up this GO term.


ProtFun2.2

In the output file of ProtFun, the "=>" indicates, which of the subcategories of each category is predicted to be true for the submitted sequence. This prediction is performed on basis of the highest information content. In the pictures below the probabilty, odds and information content (probabilty*odds) are shown separately for each category. The left y-axis is assigned to the probabilty (blue) and the information content (red line), the right y-axis is assigned to the probabilty, which is multiplied by 10 for a better perceptibility.
According to the information content the most likely functional category of the human α-galactosidase A protein is "Cell envelope". Considering our own researches, we would not come to this conclusion, but rather assign a metabolic or regulatory class, since to us known GO terms are for example "glycoside catabolic process", "negative regulation of nitric oxide biosynthetic process" and "small molecule metabolic process".
The prediction that the submitted sequence belongs to an enzyme is right.
In our opinion, it is not always the best evaluation method to decide on the basis of the information content, since the chosen enzyme class "Ligase" indeed has the highest information content, but not the highest probabilty and is, considering literature, wrong. AGAL_HUMAN clearly is a Hydrolase, which is also indicated by the high probabilty. The assigned EC number of the galactosidase is EC=3.2.1.22.
ProtFun does not predict a Gene Ontology category, since the category with the highest information content has odds lower than 1. We actually expected this category to be very unclear since the protein has very diverse cellular functions.

ProtFun uses a lot of other ressources to predict the probabilties and odds of the categories. Some of the outcomes are listed in <xr id="tab:ProtFun"/>. It is known that the α-galactosidase A protein has a 31 residues long signal peptide on position 1 to 31 which later is cleaved of. Thus the prediction of SignalP 3.0 is right.
The prediction of a propeptide cleavage site could not be confirmed in the human AGAL protein.
ProtFun infered 22 phosphorylation sites at the positions 62, 102, 201, 235, 238, 241, 276, 304, 364, 371, 405, 424, 366, 400, 86, 134, 151, 152, 173, 184, 207, 216. Phosphosite confirms phosphorylation modificatopns at S23, Y134 and H186 of which only one was predicted by NetPhos.
Uniprot claims that at the positions 139, 192, 215 and 408 there are N-linked Glycosilated sites (the last one is only a potential site). All 4 are predicted by NetNGlyc. As predicted by NetOGlyc, there are no (known) O-glycosylated sites.
See also sections Transmembrane helices and Signal peptides for further information on these two topics.

<figtable id="tab:ProtFun"> Output rendered by the individual features used by ProtFun 2.2

Feature Output summary Details
SignalP 3.0 Most likely cleavage site between pos. 31 and 32: ARA-LD Using neural networks (NN) and hidden Markov models (HMM) trained on eukaryotes
ProP 1.0 1 propeptide cleavage site predicted at position: 196 Furin-type cleavage site prediction (Arginine/Lysine residues)
TargetP 1.1 No high confidence targeting predition -
NetPhos 2.0 22 putative phosphorylation sites phosphorylation site prediction
NetOGlyc 3.1 No O-glycosylated sites predicted -
NetNGlyc 1.0 4 putative N-glycosylated sites at positions 139 192 215 408 -
TMHMM 2.0 No TM helices predicted -

</figtable>


Pfam

The Pfam sequence search revealed one significant Pfam-A match, which is shown in <xr id="fig:pfam_fam"/>.

<figtable id="fig:pfam_fam"> Pfam-A match

Family Description Entry xtype Clan Envelope Start Envelope End Alignment Start Alignment End HMM From HMM To Bit score E-value
Melibiase Melibiase Family CL0058 33 149 40 146 41 140 50.0 1.5e-13

</figtable>


<figure id="fig:AGAL">

The protein encoded by the Fabry-associated GLA gene: α-Galactosidase A

</figure>


The alpha-galactosidase A protein is, according to PFAM, in the family of the Melibiases (see <xr id="tab:Pfam_Melibiose"/>). The Pfam PF02065 family itself is refered to as "Glycoside hydrolase family 27" and "Glycoside hydrolase family 36", the AGAL_HUMAN falls into the first category. Their common characteristic is that the members of this family are glycoside hydrolases (EC 3.2.1.). The AGAL enzyme catalyzes the hydrolisis of the disaccharide Melibiose (D-Gal-α(1→6)-D-Glc) into its two components galactose and glucose.

<figtable id="tab:Pfam_Melibiose"> Melibiase Identifiers

Melibiase Identifiers
Symbol Melibiase
Pfam PF02065
Pfam clan CL0058
InterPro IPR000111
SCOP 1ktc
SUPERFAMILY 1ktc
CAZy GH27

</figtable>

The alignment of positions 40 - 146 of the AGAL_HUMAN protein sequence and the matching HMM used in this prediction, is shown in <xr id="fig:pfam_ali" />. According to the color code, indicating the degree of confidence of each aligned position, there is an overall very good agreement of the Hidden Markov Model and the sequence, except for the residues 17-46. Checking our background knowledge and the Uniprot database we could not find a very interesting or abnormal region here, but the signal peptide cleavage site and a beta strand at position 42-46.
Our query protein belongs to a rather large Clan, the Glyco_hydro_tim Clan (CL0058), which includes 4 CAZy-Clans (GH-A, GH-D, GH-H and GH-K). They main attribute of all the included glycosyl hydrolase enzymes is the hold of a a TIM barrel fold (eight α-helices and eight parallel β-strands that alternate along the peptide backbone source). This fold (residue 31 - 324) can be well seen in <xr id="fig:AGAL"/>, where the 3D structure of α-Galactosidase A is depicted.
The InterPro protein sequence analysis and classification assigns the enzyme to the "IPR000111 Glycoside hydrolase, clan GH-D" family. As mentioned before, GH-D is part of the Glyco_hydro_tim Clan and thus it is not surprising, that the description is almost the same. According to InterPro, there are 6 IPR000111 family members in the human body.
The Structural Classification of Proteins (SCOP) finds two domains, an Amylase, catalytic domain (c.1.8.1, residues 32 - 323) and an alpha-Amylases, C-terminal beta-sheet domain (b.71.1.1, residues 324 - 421). The catalytic domain is an alpha and beta protein (a/b) with TIM beta/alpha-barrel fold, which is consonant with the affiliation to the Glyco_hydro_tim Clan. The C-terminal beta-sheet domain is an all beta protein, the fold and superfamily is Glycosyl hydrolase domain. This is also not surprising to us.
AGAL is in the CAZy GH27 (Glycoside Hydrolase Family 27) family, together with a-N-acetylgalactosaminidase (EC 3.2.1.49; as mentioned before, used for enzyme replacement therapy), isomalto-dextranase(EC 3.2.1.94) and b-L-arabinopyranosidase (EC 3.2.1.88). See http://www.cazypedia.org/index.php/Glycoside_Hydrolase_Family_27 TODO

<figure id="fig:pfam_ali">

Melibiase - alignment region residues 40 - 146

</figure>


Other programs and ressources

A simple internet search for "protein sequence prediction" reveals a confusingly high number of different ressources and databases. Clicking on the first entry leads to Rostlab's Predict protein. This program seems to predict nearly anything:

  1. Multiple sequence alignment
  2. ProSite sequence motifs
  3. low-complexity retions (SEG)
  4. Nuclear localisation signals
  5. and predictions of
    • secondary structure
    • solvent accessibility
    • globular regions
    • transmembrane helices
    • coiled-coil regions
    • structural switch regions
    • B-value
    • disordered regions
    • intra-residue contacts
    • protein protein and protein/DNA binding sites
    • sub-cellular localization
    • domain assignment
    • beta barrels
    • cysteine predictions and disulphide bridges

Performing a search reveals that the tool gives you an overview of all the above mentioned features. On a first glance it seemed rather confusing, due to the vast amount of informations provided. Anyhow I believe giving the output some more attentiveness, it is a good tool to gather a lot of information in one place. A very good feature is that the output is presented in many different ways which can be switched at any time. A quick look at the properties we examined in the task, shows, that Predict Protein probably does not distinguish very well between signal peptides and transmembrane helices, since for the α-Galactosidase A it predicts a TM, where there actually is signal peptide (position 11-28; see section Transmembrane helices and Signal peptides)
Searching the same term again at the NCBI homepage (see search) reveals a huge mass of reviews on that topic, mainly on predicting secondary structure and function out of the protein sequence, but also on less common topics like predicting the protein solubility (see here and here) , or more recently emerged whole genome sequence approaches (see).
A really different topic was brought up by Smialowski P, Martin-Galiano AJ, Cox J and Frishman D. who give an overview of techniques to predict experimental properties of proteins from sequence (see) and thus predict experimental success in cloning, expression, soluble expression, purification and crystallization
Summarizing, it is very hard to get along in the vast amount of prediction tools and methods and one can spend a lot of time on finding out which one fits best to the requirements of one's topic.