Difference between revisions of "Task 4: Homology based structure predictions"
(→Homology modelling with Modeller) |
(→Homology modelling with Modeller) |
||
Line 187: | Line 187: | ||
* The used fix of the python installation is described at the [[resource_software|software section]]. |
* The used fix of the python installation is described at the [[resource_software|software section]]. |
||
* The Modeller modules were still not importable by Python, that is why it was necessary to reinstall Modeller. The steps for this are described in [[resource_software|software section]]. |
* The Modeller modules were still not importable by Python, that is why it was necessary to reinstall Modeller. The steps for this are described in [[resource_software|software section]]. |
||
+ | |||
+ | After all this stuff we tried to write a alignment-file with the hhsearch alignments and a python-file (according to the example <code>/apps/modeller9.9/examples/automodel/model-default.py</code>. But modeller seems to be very sensible due to missing acids in the coordinate section of the pdb, which are of course mentioned in the sequence used in the alignment. |
||
+ | |||
+ | To avoid this problem there are at least two possibilities: |
||
+ | * Repair the pdb by the script repairPDB. Map the sequence in the alignment on the sequence in the coordinate section of the used pdb (a lot of work - need some kind of alignment and a pdb-parser... both is not trivial) |
||
+ | * The other possibility is to let Modeller create the alignment on the basis of the repaired PDB (with repairPDB). |
||
+ | |||
+ | We chose the second option, which is described in the Modeller tutorial (see [http://salilab.org/modeller/tutorial/basic.html basic modeller tutorial]. |
||
=== Homology modelling with Swissmodel === |
=== Homology modelling with Swissmodel === |
Revision as of 14:03, 8 June 2011
Task description
The full description of this task can be found here.
Calculation of models
Overview of available homologous structures
Search
We used hhsearch with the standard parameter to find homologous structures of our protein. The following command was executed:
- ./hhsearch -i reference_pah_aa.fasta -d pdb70.db -b 500 -o hhsearch.out
We received the following hits:
No. | PDB ID | Description | Prob | E-Value | P-Value | Score | SS | Cols | Query HMM | Template HMM | Residues | Sequence Identity |
---|---|---|---|---|---|---|---|---|---|---|---|---|
1 | 1phz_A | Protein (phenylalanine | 1 | 1 | 1 | 1084.4 | 0 | 429 | 1-429 | 1-429 | (429) | 92% |
2 | 1j8u_A | Phenylalanine-4-hydroxy | 1 | 1 | 1 | 894.5 | 0 | 325 | 103-427 | 1-325 | (325) | 100% |
3 | 1toh_A | Tyroh tyrosine hydroxy | 1 | 1 | 1 | 890.7 | 0 | 342 | 111-452 | 2-343 | (343) | 60% |
4 | 1mlw_A | Tryptophan 5-monooxygen | 1 | 1 | 1 | 804.2 | 0 | 300 | 116-415 | 2-301 | (301) | 66% |
5 | 1ltz_A | Phenylalanine-4-hydroxy | 1 | 1 | 1 | 504.9 | 0 | 265 | 144-414 | 2-269 | (297) | 30% |
6 | 2v27_A | Phenylalanine hydroxyla | 1 | 1 | 1 | 471.1 | 0 | 254 | 167-424 | 4-271 | (275) | 30% |
7 | 2qmx_A | Prephenate dehydratase; | 1 | 1 | 1 | 70.0 | 0 | 53 | 33-85 | 199-251 | (283) | 40% |
8 | 2qmw_A | PDT prephenate dehydra | 1 | 1 | 1 | 66.1 | 0 | 51 | 35-85 | 190-240 | (267) | 37% |
9 | 3luy_A | Probable chorismate mut | 1 | 1 | 1 | 66.0 | 0 | 53 | 33-85 | 207-259 | (329) | 28% |
10 | 1y7p_A | Hypothetical protein AF | 1 | 1 | 1 | 19.9 | 0 | 38 | 36-73 | 6-43 | (223) | 16% |
Template structure selection
We selected the following structures as our template structures:
- > 60% sequence identity: 1phz
- > 40% sequence identity: 1toh
- < 40% sequence identity: 1ltz
Homology modelling with Modeller
Modeller uses two types of files to be run. The first one contains the used alignment in the PIR-format (see PIR FORMAT), the second one is a python script, which tells Modeller which steps it has to perform. There are some examples in /apps/modeller9.9/examples/automodel
. But python seems to be falsely configured on the virtual machines (at least the linux virtual machine).
- The used fix of the python installation is described at the software section.
- The Modeller modules were still not importable by Python, that is why it was necessary to reinstall Modeller. The steps for this are described in software section.
After all this stuff we tried to write a alignment-file with the hhsearch alignments and a python-file (according to the example /apps/modeller9.9/examples/automodel/model-default.py
. But modeller seems to be very sensible due to missing acids in the coordinate section of the pdb, which are of course mentioned in the sequence used in the alignment.
To avoid this problem there are at least two possibilities:
- Repair the pdb by the script repairPDB. Map the sequence in the alignment on the sequence in the coordinate section of the used pdb (a lot of work - need some kind of alignment and a pdb-parser... both is not trivial)
- The other possibility is to let Modeller create the alignment on the basis of the repaired PDB (with repairPDB).
We chose the second option, which is described in the Modeller tutorial (see basic modeller tutorial.
Homology modelling with Swissmodel
Standard workflow
The standard workflow of Swissmodel is the automated mode. For this mode only the UniProt accession number or the amino acid sequence of the target protein is required. As an optional parameter it is possible to enter the template structure as well. However, if this field is left blank Swissmodel will search automatically for a suitable template.
The input for all three models is as follows:
Homology modelling with iTasser
Evaluation of the calculated models
Selection of the reference structures
We had the following choice of reference structures for PAH:
Entry | Method | Resolution (A) | Chain | Positions |
---|---|---|---|---|
1DMW | X-Ray | 2.00 | A | 118-424 |
1J8T | X-Ray | 1.70 | A | 103-427 |
1J8U | X-Ray | 1.50 | A | 103-427 |
1KW0 | X-Ray | 2.50 | A | 103-427 |
1LRM | X-Ray | 2.10 | A | 103-427 |
1MMK | X-Ray | 2.00 | A | 103-427 |
1MMT | X-Ray | 2.00 | A | 103-427 |
1PAH | X-Ray | 2.00 | A | 117-424 |
1TDW | X-Ray | 2.10 | A | 117-424 |
1TG2 | X-Ray | 2.20 | A | 117-424 |
2PAH | X-Ray | 3.10 | A/B | 118-452 |
3PAH | X-Ray | 2.00 | A | 117-424 |
4PAH | X-Ray | 2.00 | A | 117-424 |
5PAH | X-Ray | 2.10 | A | 117-424 |
6PAH | X-Ray | 2.15 | A | 117-424 |
All these structures have in common that they did not solve the structure of the whole PAH protein. In addition, there is no complete true apo structure available either. All structures have at least a Fe2+ atom bound. So we defined these structures as our apo structure.
Finally, we decided to select 1J8T (apo) and 1J8U (complexed). As mentioned before our apo structure has complexed Fe2+ and our complexed structure is complexed with Fe2+ and BH4 (5,6,7,8-TETRAHYDROBIOPTERIN). The reason for our decision was that both structures are solved from the same group which somehow guaranties a more consistent methodology as if we had selected structures from two different groups. Another reason is the resolution, both structures are the two with the best resolved resolution which is 1.5 Angstrom and 1.7 Angstrom for 1J8U and 1J8T respectively. Finally for more easy comparison, both structures include the same range of amino acids which is from 103 to 427.
Numeric evaluation of the calculated models
Modeller
Swissmodel
Automated Mode: Modelling with template structure 1phz_A (>60%)
QMEAN Z-Score: -0.828
QMEAN4 global scores:
QMEANscore4 | Estimated absolute model quality | Score components |
---|---|---|
0.715 |
Local scores:
Coloring by residue error | Residue error plot |
---|---|
Global scores: QMEAN4:
Scoring function term | Raw score | Z-score |
---|---|---|
C_beta interaction energy | -157.98 | -0.16 |
All-atom pairwise energy | -12503.57 | -0.1 |
Solvation energy | -50.66 | 1.01 |
Torsion angle energy | -78.77 | -1.48 |
QMEAN4 score | 0.715 | -0.83 |
Local Model Quality Estimation: Anolea / QMEAN / Gromos:
Automated Mode: Modelling with template structure 1toh_A (>40%)
QMEAN Z-Score: -2.745
QMEAN4 global scores:
QMEANscore4 | Estimated absolute model quality | Score components |
---|---|---|
0.604 |
Local scores:
Coloring by residue error | Residue error plot |
---|---|
Global scores: QMEAN4:
Scoring function term | Raw score | Z-score |
---|---|---|
C_beta interaction energy | -78.67 | -1.16 |
All-atom pairwise energy | -7899.47 | -0.89 |
Solvation energy | -20.56 | -1.3 |
Torsion angle energy | -47.34 | -2.22 |
QMEAN4 score | 0.604 | -2.74 |
Local Model Quality Estimation: Anolea / QMEAN / Gromos:
Automated Mode: Modelling with template structure 1ltz_A (<40%)
QMEAN Z-Score: -4.282
QMEAN4 global scores:
QMEANscore4 | Estimated absolute model quality | Score components |
---|---|---|
0.47 |
Local scores:
Coloring by residue error | Residue error plot |
---|---|
Global scores: QMEAN4:
Scoring function term | Raw score | Z-score |
---|---|---|
C_beta interaction energy | -40.25 | -2.1 |
All-atom pairwise energy | -3528.81 | -2.29 |
Solvation energy | -15.22 | -1.18 |
Torsion angle energy | -4.99 | -3.78 |
QMEAN4 score | 0.47 | -4.28 |
Local Model Quality Estimation: Anolea / QMEAN / Gromos:
iTasser
Comparison to experimental structure
Modeller
Swissmodel
To calculate the C-alpha RMSD we used DaliLite.
To calculate the TM-Score we used the TM-score webservice from the University of Michigan.
To calculate the RMSD of the 6A radius of the catalytic center we had to first identify the catalytic center. We defined the center position of the catalytic side as the position where our Fe2+ atom is. With the position in hand we now have to extract the residues in a 6A radius around this Fe2+ atom. In order to do so we executed the following steps:
- We opened the complexed or apo structure and one of the modeled structures with Pymol.
- Then we aligned both structures to each other
- Then we selected the Fe2 atom of the apo/complexed structure and expanded this selection by 6A, residue
- Then we extracted the selected residues into two objects each object contains only the residues of either the apo/complexed structure or the modeled structure
- Then we saved both objects in seperate PDB structures
- Now we used the rms.pl script to calculate the all atom RMSD with the following command "./rms.pl -out all first.pdb second.pdb"
Standard Workflow
Template Structure | Compared To | Apo/Complexed | C-alpha RMSD | TM score | All Atoms RMSD, 6A |
---|---|---|---|---|---|
1PHZ Chain: A | 1J8T | Apo | 0.9 | 0.7400 | 0.5162 |
1PHZ Chain: A | 1J8U | Complexed | 0.9 | 0.7408 | 0.5154 |
1TOH Chain: A | 1J8T | Apo | 1.3 | 0.8889 | 0.4616 |
1TOH Chain: A | 1J8U | Complexed | 1.2 | 0.8894 | 0.3361 |
1LTZ Chain: A | 1J8T | Apo | 2.3 | 0.8816 | 0.9225 |
1LTZ Chain: A | 1J8U | Complexed | 2.3 | 0.8814 | 0.9208 |
Adjusted Alignment Workflow
Template Structure | Compared To | Apo/Complexed | C-alpha RMSD | TM score | All Atoms RMSD, 6A |
---|---|---|---|---|---|
1PHZ Chain: A | 1J8T | Apo | |||
1PHZ Chain: A | 1J8U | Complexed | |||
1TOH Chain: A | 1J8T | Apo | |||
1TOH Chain: A | 1J8U | Complexed | |||
1LTZ Chain: A | 1J8T | Apo | |||
1LTZ Chain: A | 1J8U | Complexed |