AmphipaSeeK help
A brief introduction to AmhpipaSeeK
AmphipaSeeK is a prediction method of amphipathic in-plane membrane anchors
(IPM anchors) in protein sequences. IPM anchoring is a particular type of membrane interaction
found as well as in monotopic proteins (e.g. NonStructural protein 5A of the Hepatitis C
virus [1]) and in polytopic proteins (e.g. gp41 from the Human Immunodeficiency Virus [2]).
AmphipaSeeK has been initally developed with a set of 21 monotopic membrane proteins including an
experimentally characterized IPM anchor. The set is currently extended to IPM segments found in
transmembrane proteins. To date, 7 transmembrane proteins are included in the data set.
[ ! ] AmphipaSeeK is not a prediction method of transmembrane
segments or monotopic proteins. It only tells if a protein sequence potentially contains an IPM anchor [ ! ]
AmphipaSeeK uses a SVM classifier based on Gaussian kernel. The classification
process requires a Gram matrix derived from a substitution matrix (currently : PHAT
[3]). This part of AmphipaSeeK code come from the M-SVM source code
distributed by Yann Guermeur under the GNU public License. M-SVM is a Multiclass Support Vector Machine,
for details see [4]. Additionally, AmphipaSeeK automatically generates a multiple alignment to compute
a weighted average prediction. Homologous to your query sequence are retrieve from the UniProt database.
If no homologous sequence is retrived, the prediction will be computed on your query only.
The computation of the average prediction uses the Squid-1.9g library
distributed by Stephen Eddy under the GNU Public License.
|
If you use AmphipaSeeK in a publication, please cite :
Sapay N, Guermeur Y, and Deléage G (2006) prediction of amphipathic in-plane membrane anchors
in monotopic proteins using a SVM classifier, BMC Bioinformatics, 7:255.
|
References :
- Sapay N, Montserret R, Chipot C, Brass V, Deléage G, Moradpour D, and Penin F (2006) NMR
structure and molecular dynamics of the in-plane membrane anchor of the monotopic NonStructural protein 5A
(NS5A) from the Bovine Viral Diarrhea Virus (BVDV), Biochemistry, 45(7):2221-2233.
- Schibli DJ, Montelaro RC, and Vogel HJ (2001) The membrane-proximal tryptophan-rich region of
the HIV glycoprotein, gp41, forms a well-defined helix in dodecylphosphocholine micelles, Biochemistry,
40(32):9570-9578.
- Ng PC, Henikoff JG and Henikoff S (2001) PHAT: a transmembrane-specific substitution matrix.
predicted hydrophobic and transmembrane, Bioinformatics, 16(9):760-766.
- Guermeur Y, Lifchitz A, and Vert R. (2004) A kernel for protein secondary structure prediction,
In Schölkopf B, Tsuda K and Vert JP, editors, Kernel Methods in Computational Biology.
List of known IPM anchored proteins currently used in AmphipaSeeK
List of transmembrane proteins including a known IPM anchor currently used in AmphipaSeeK
Availability in NPS@
AmphipaSeeK is available :
Paste your protein sequence and submit! By default, AmphipaSeeK will compute a prediction as
sensitive as possible, based on the set of 21 + 7 membrane proteins. This corresponds to the option
[1] . For more specific prediction, you can try option [3]. If you really the most specific
efficiency, try option [6]. Supplemental options are provided to help you to analyze AmphipaSeeK results,
including :
- the secondary structure prediction to looking for putative α-helical segments.
- the hydrophobic moment μH calculation to measures the amphipathy along the sequence.
- a basic smoothing function to avoid predicted IPM anchors made of a single residue.
[ ! ] Analyses options are not used by AmphipaSeeK
to compute a predictions. They are just a help [ ! ]
Parameters
-
SVM efficiency :
choice of the SVM training set, specificity and sensitivity.
Available options correspond to 6 different SVM training parameterizations giving 6 different performance.
For specific predictions, try option [3]. If you prefer for sensitive predictions, the option [1]
is more suitable. Efficiencies of each options are summurized in table 1 :
| Option |
Training Set |
Accuracy |
Sensitivity |
Specificity |
Positive Predictive Value |
Pearson-Matthews Coefficient |
| [1] |
30 monotopic |
94.2 |
31.0 (low) |
99.5 (high) |
82.9 (high) |
0.49 |
| [2] |
30 monotopic |
93.9 |
33.3 (high) |
99.0 (low) |
73.8 (low) |
0.47 |
| [3]a |
30 monotopic + 8 transmembrane proteins |
92.9 |
23.1 (low) |
99.3 (high) |
74.6 (high) |
0.39 |
| [4]a |
30 monotopic + 8 transmembrane proteins |
92.5 |
26.0 (high) |
98.6 (low) |
62.3 (low) |
0.37 |
Table 1. Efficiency associated to the 4 SVM parameterizations.
a Choices [3] and [4] are still under experiments.
-
Gaussian kernel bandwidth : the SVM gaussian kernel bandwith. Default is an
optimal value depending on the chosen efficiency. Please read the publication of
AmphipaSeeK for details
-
Use prediction smoothing : to smooth the AmphipaSeeK score after the computation.
This is useful to avoid predicted IPM anchors including a single residue.
-
Smoothing window size : size of the sliding window used to smooth predictions.
-
Smoothing factor : factor use to weight the smoothing. Increase it to diminish the smoothing.
-
Output width : size of the blocks displayed in the AmphipaSeeK output file.
Default is 80 residues.
-
μH window size : size of the sliding window used in the hydrophobic moment μH
computation.
[ ! ] The μH calculation is a supplemental option and is
not used by AmphipaSeeK to predict IPM anchors [ ! ]
-
μH periodicity : periodicity used in theμH calculation. Chose 100°
for a canonical α-helix or 180° for a canonical β-strand.
[ ! ] The μH calculation is a supplemental option and is
not used by AmphipaSeeK to predict IPM anchors [ ! ]
-
Show predicted secondary structure : to display the predicted secondary
structure (using the MLRC method) with the AmphipaSeeK results.
[ ! ] The secondary structure prediction is a supplemental option and is
not used by AmphipaSeeK to predict IPM anchors [ ! ]
-
NPS@ AmphipaSeeK output example
The output is divided into 1 HTML page and 4 text files.
- HTML page
- Part 1/3. prediction computed on your query sequence in text format
10 20 30 40 50.......... sequence position
| | | | |
SGSWLRDIWDWICEVLSDFKTWLKAKLMPQLPGIPFVSCQRGYRGVWRGD.......... protein sequence
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA .......... AmphipaSeeK prediction : A = IPM anchor
cchhhhhhhhhhhhhhhhhhhhhhhhccccccccceeeccccceeeeecc.......... predicted secondary structure
33434455555444444344322222222332111222334455554442.......... amphipathy : from 0=low to 5=high amphipathy
based on the sequence-average μH
- 0 : μH > average - 2sd
- 1 : average - 2sd ≥ μH > average - 1sd
- 2 : average - 1sd ≥ μH > average
- 3 : average ≥ μH > average + 1sd
- 4 : average + 1sd ≥ μH > average + 2sd
- 5 : average + 2sd ≥ μH
- Part 2/3. Plot of the AmphipaSeeK score : score as a function of the sequence position. predicted IPM anchors have a score upper than 0.00.
- Part 3/3. Links to AmphipaSeeK output files :
result file............................... AmphipaSeeK results in a nice text format
result score.............................. score computed by AmphipaSeeK in an aligned text format
used alignment............................ multiple alignment used to compute an average score (in CLUSTALW format)
predicted secondary structure consensus... predicted secondary structure in NPS@ format
- Text file 1/4. AmphipaSeeK results
- Text file 2/4. AmphipaSeeK scores
- Part 1/2. Score computed on your query sequence :
>JohnDoe ............................................................. sequence header
# rank residue score topology structII <muH> amphipathy... score associated to each
1 S 0.028 A c 0.438 3... information associated to the sequence position
2 G 0.056 A c 0.415 3 (membrane topology, secondary structure, μH, ...)
3 S 0.198 A h 0.457 4
...
- Part 2/2. prediction computed on the aligned sequences if available : Same remark as in file 1,
part 3.
- Text file 3/4. Multiple alignment : Same remark as in file 1, part 3.
- Text file 4/4. predicted secondary structure : predicted secondary structure of your query by SOPMA.
[ ! ] The secondary structure prediction is a supplemental option.
It is not used by AmphipaSeeK to predict IPM anchors [ ! ]
References