Gaucher Disease: Task 05 - Homology Modelling
<css>
table.colBasic2 { margin-left: auto; margin-right: auto; border: 1px solid black; border-collapse:collapse; }
.colBasic2 th,td { padding: 3px; border: 1px solid black; }
.colBasic2 td { text-align:left; }
/* for orange try #ff7f00 and #ffaa56 for blue try #005fbf and #aad4ff
maria's style blue: #adceff grey: #efefef
- /
.colBasic2 tr th { background-color:#efefef; color: black;} .colBasic2 tr:first-child th { background-color:#adceff; color:black;}
</css>
Contents
Calculation of models
For information on the selected structure set see lab journal.
We created two single target models of our protein sequence, P04062: one with a high sequence identity target, 2XWD_A, and one with a low sequence identity target, 2WNW_A, with each one of the three tools: Modeller, Swiss-Model and iTasser.
For Modeller we additionally executed multiple target modeling mode. In the multiple target modeling mode, Modeller first aligns the user selected templates, then adds the target to the MSA, which is finally used for modeling. We tried the following template combinations:
- close homologues (> 60% sequence identity): all four (3KE0_A, 2XWD_A, 2WKL_A and 2NSX_A)
- distant homologues (< 30% sequence identity): all three (2WNW_A, 1VFF_A and 3II1_A)
- close and distant homologues: 2XWD_A and 2WNW_A
Evaluation of models
In the following we present the results of the created models. We compare the models to two reference structures - 1OGS_A and 2V3E_B - by superimposing and calculating C_alpha RMSD by Pymol as well as all atom RMSD, GDT-TS-score and TM-scoreby TMscore tool. In addition, we compare main scores given by the modelling programs.
- Select one apo and one complex structure if there are several experimental structures -> document your choice of reference
We selected 1OGS_A for consistency reasons (as we already used it in the task 04). Moreover, as there are many reference structures for glucocerebrosidase, we chose another structure for comparison: 2V3E_B. Both structure have resolution of 2.0, however 2V3E was resolved at neutral pH, whereas 1OGS had an acidic pH-value of 4.5. (Much more detailed information about the structures is given in the lab journal of task 09.)
For brief explanation of the execution of the programs and calculation of the scores, please consult lab journal.
Modeller
<figtable id="single_templates">
| High sequence identity | ||||
|---|---|---|---|---|
| Template | 2XWD_A | |||
| Alignment method | malign | salign | ||
| DOPE score | -64821.74609 | -64821.746094 | ||
| Reference targets | 1OGS_A | 2V3E_B | 1OGS_A | 2V3E_B |
| C_alpha RMSD (# atoms pairs) | 0.284 (408) | 0.194 (444) | 0.284 (408) | 0.194 (444) |
| RMSD (of 497 common residues) | 23.987 | 24.032 | 23.987 | 24.032 |
| TM score | 0.2160 | 0.2167 | 0.2160 | 0.2167 |
| GDT-TS score (%) | 0.0664 | 0.0669 | 0.0664 | 0.0669 |
| Pymol visualization | ||||
| Low sequence identity | ||||
| Template | 2WNW_A | |||
| Alignment method | malign | salign | ||
| DOPE score | -57316.70703 | -53548.45703 | ||
| Reference targets | 1OGS_A | 2V3E_B | 1OGS_A | 2V3E_B |
| C_alpha RMSD (# atoms pairs) | 1.217 (340) | 1.116 (336) | 1.523 (337) | 1.523 (337) |
| RMSD (of 497 common residues) | 22.979 | 22.980 | 22.972 | 22.949 |
| TM score | 0.2301 | 0.2307 | 0.2257 | 0.2249 |
| GDT-TS score (%) | 0.0659 | 0.0664 | 0.0634 | 0.0649 |
| Pymol visualization | ||||
</figtable>
<figtable id="multiple_templates">
| High sequence identity | ||
|---|---|---|
| Templates | 3KE0_A, 2XWD_A, 2WKL_A, 2NSX_A | |
| DOPE score | NA | |
| Reference targets | 1OGS_A | 2V3E_B |
| C_alpha RMSD (# atoms pairs) | 0.272 (439) | 0.337 (413) |
| RMSD (of 497 common residues) | 24.370 | 24.402 |
| TM score | 0.2173 | 0.2175 |
| GDT-TS score (%) | 0.0664 | 0.0674 |
| Pymol visualization | ||
| Low sequence identity | ||
| Templates | 2WNW_A, 1VFF_A, 3II1_A | |
| DOPE score | NA | |
| Reference targets | 1OGS_A | 2V3E_B |
| C_alpha RMSD (# atoms pairs) | 20.935 (471) | 21.012 (473) |
| RMSD (of 497 common residues) | 22.759 | 22.875 |
| TM score | 0.1931 | 0.1924 |
| GDT-TS score (%) | 0.0508 | 0.0523 |
| Pymol visualization | ||
| Mixed sequence identity | ||
| Templates | 2XWD_A, 2WNW_A | |
| DOPE score | NA | |
| Reference targets | 1OGS_A | 2V3E_B |
| C_alpha RMSD (# atoms pairs) | 0.328 (418) | 0.228 (450) |
| RMSD (of 497 common residues) | 23.625 | 23.659 |
| TM score | 0.2170 | 0.2175 |
| GDT-TS score (%) | 0.0664 | 0.0669 |
| Pymol visualization | ||
</figtable>
DOPE score is an energy score, this means the model with the lowest DOPE score is the best. (Modeller scores are also explained in the tutorial.)
Swiss-Model
<figtable id="swiss-model">
| High sequence identity | ||
|---|---|---|
| Template | 2XWD_A | |
| QMEAN Z-score | -1.37 | |
| Reference targets | 1OGS_A | 2V3E_B |
| C_alpha RMSD (# atoms pairs) | 0.305 (407) | 0.198 (447) |
| RMSD (of # common residues) | 23.688 (458) | 23.737 (458) |
| TM score | 0.2137 | 0.2143 |
| GDT-TS score (%) | 0.0654 | 0.0664 |
| Pymol visualization | ||
| Low sequence identity | ||
| Template | 2WNW_A | |
| QMEAN Z-score | -4.1 | |
| Reference targets | 1OGS_A | 2V3E_B |
| C_alpha RMSD (# atoms pairs) | 1.105 (368) | 1.032 (372) |
| RMSD (of # common residues) | 22.058 (422) | 22.022 (422) |
| TM score | 0.2129 | 0.2138 |
| GDT-TS score (%) | 0.0634 | 0.0653 |
| Pymol visualization | ||
</figtable>
Swiss-model's QMEAN4 score is explained on the result page:
The QMEAN4 score is a composite score consisting of a linear combination of 4 statistical potential terms:
- C_beta interaction energy
- All-atom pairwise energy
- Solvation energy
- Torsion angle energy
Its meaning is estimated model reliability (between 0-1). The pseudo-energies of the contributing terms are calculated together with their Z-scores with respect to scores obtained for high-resolution experimental structures of similar size solved by X-ray crystallography.
iTasser
<figtable id="itasser">
| High sequence identity | ||
|---|---|---|
| Template | 2XWD_A | |
| C-score | -0.33 | |
| Predicted RMSD | 8.2±4.5 | |
| Predicted TM score | 0.67±0.13 | |
| Reference targets | 1OGS_A | 2V3E_B |
| C_alpha RMSD (# atoms pairs) | 0.692 (461) | 0.755 (467) |
| RMSD (of # common residues) | 24.427 (497) | 24.487 (497) |
| TM score | 0.2171 | 0.2175 |
| GDT-TS score (%) | 0.0674 | 0.0674 |
| Pymol visualization | ||
| Low sequence identity | ||
| Template | 2WNW_A | |
| C-score | -0.40 | |
| Predicted RMSD | 8.4±4.5 | |
| Predicted TM score | 0.66±0.13 | |
| Reference targets | 1OGS_A | 2V3E_B |
| C_alpha RMSD (# atoms pairs) | 1.232 (431) | 1.185 (439) |
| RMSD (of # common residues) | 24.087 (497) | 24.147 (497) |
| TM score | 0.2196 | 0.2208 |
| GDT-TS score (%) | 0.0694 | 0.0699 |
| Pymol visualization | ||
</figtable>
iTasser scores are explained on the results page:
C-score is a confidence score for estimating the quality of predicted models by I-TASSER. It is calculated based on the significance of threading template alignments and the convergence parameters of the structure assembly simulations. C-score is typically in the range of [-5,2], where a C-score of higher value signifies a model with a high confidence and vice-versa.
TM-score and RMSD are known standards for measuring structural similarity between two structures which are usually used to measure the accuracy of structure modeling when the native structure is known. In case the native structure is not known, it becomes necessary to predict the quality of the modeling prediction, i.e. what is the distance between the predicted model and the native structures? To answer this question, we tried predicted the TM-score and RMSD of the predicted models relative the native structures based on the C-score.
TM-score is a recently proposed scale for measuring the structural similarity between two structures (see Zhang and Skolnick, Scoring function for automated assessment of protein structure template quality, Proteins, 2004 57: 702-710). The purpose of proposing TM-score is to solve the problem of RMSD which is sensitive to the local error. Because RMSD is an average distance of all residue pairs in two structures, a local error (e.g. a misorientation of the tail) will araise a big RMSD value although the global topology is correct. However, in TM-score, the small distance is weighted stronger than the big distance which makes the score insensitive to the local modeling error. A TM-score >0.5 indicates a model of correct topology and a TM-score<0.17 means a random similarity. This cutoff does not depends on the protein length.
C_alpha RMSD summary
In <xr id="models_RMSD"/> C_alpha RMSD computed by Pymol with 1OGS_A as reference are summarized.
<figtable id="models_RMSD">
| Program | Templates | ||||
|---|---|---|---|---|---|
| High PIDE (2XWD) | low PIDE (2WNW) | high PIDE (4 temp.) | low PIDE (3 temp.) | mixed PIDE (2XWD & 2WNW) | |
| Modeller | 0.284 (408) | 1.523 (337) | 0.272 (439) | 20.935 (471) | 0.328 (418) |
| Swiss-Model | 0.305 (407) | 1.105 (368) | |||
| iTasser | 0.692 (461) | 1.232 (431) | |||
</figtable>
All atom RMSD calculation near the binding sites (extra diligence task)
Defining a radius of 6 Angstrom around the ligands of 1OGS_A and 2V3E_B (the catalytic centers / binding sites), we calculated the all atom RMSD in that region between the best model - Modeller multiple templates with high PIDE - and the reference structures. The structure 1OGS_A has one ligand, NAG, which bind to a loop on the outer side of the protein. The structure 2V3E_B has multiple ligands that bind to three different binding sites:
- the same binding site as in 1OGS_A: 2 NAG molecules, BMA, FUC
- a loop on the other outer side: third NAG molecule
- a groove in the center of a protein: NND
<figure id="contact_ras" >
</figure>
The resulting RMSD value are summarized in the table <xr id="binding_site_rmsd"/> and the Pymol visualization is shown in <xr id="contact_ras"/>.
<figtable id="binding_site_rmsd">
| Reference | 1OGS_A | 2V3E_B |
|---|---|---|
| RMSD <6 of ligand(s) (#atoms) | 0.266 (29) | 0.469 (108) |
</figtable>
The all atom RMSD values for the binding sites are pretty low (0.266 and 0.469), much lower then the respective all atom RMSD values for the entire molecules (24.37 and 24.402) and resemble the C-alpha RMSD values for the entire molecules (0.272 and 0.337). The RMSD for the binding sites of 2V3E_B is higher, then for the binding site of 1OGS_A, because in 2V3E_B all three binding sites are included and the number of atom pairs is higher (108 vs. 29). Overall, Modeller aligned the binding site residue well.
Discussion
All-atom vs. C-alpha RMSD
In both cases of RMSD, some distant atoms pairs beyond some threshold were omitted from the calculation. All-atom RMSD is relatively high and sometimes even higher for models created with low sequence identity templates. Pymol C-alpha RMSD is much lower and is lower for models created with high sequence identity, which also look good aligned with the reference structures. Therefore, in the following usage of "RMSD" we always regard the C-alpha RMSD.
Reference structures
In most of the cases, there are almost no differences in the scores when using 1OGS_A or 2V3E_B as a reference template structure. In the following, we regard RMSD values compared to the structure 1OGS, but the trend applied also with 2V3E_B as reference.
How do the RMSD and GDT correlate? Is one score more helpful in finding meaningful models?
The lower the C_alpha RMSD, the higher the GDT-TS score. This means, that a good quality model has a high GDT score. It ranges between 0 and 1. However, GDT scores we received range only between 0.0508 and 0.0699.
Do you see any correlation between the quality scores provided by the modelling tools and the RMSD/GDT?
Modeller DOPE score becomes lower (=better energy) with falling RMSD and rising GDT score. (DOPE was not given the multiple structure models results.)
Swiss-Model QMEAN Z-score becomes more negative for higher RMSD and lower GDT score. It means that models having a higher (or smaller negative number) QMEAN Z-score are better. The same correlation applies for iTasser C-score. ("C-score is typically in the range of [-5,2], where a C-score of higher value signifies a model with a high confidence and vice-versa.")
Is any method systematically better at predicting the structure and does this depend on the similarity of the template?
Looking again at the C_alpha RMSD of all models (see <xr id="models_RMSD"/>), Modeller produced the best two models: using multiple high PIDE templates (RMSD=0.272) and the high PIDE 2XWD (RMSD=0.284) as a template. The model using mixed PIDE templates war also relatively good (RMSD=0.328). The low PIDE template (2WNW) yielded worse results (RMSD=1.523) and using 3 low PIDE templates led to RMSD of 20.935.
Swiss-model produced the third best alignment (RMSD=0.305), also using the high PIDE template. Remarkably, it was the best method using the low PIDE template (RMSD=1.105), the superposed model looks really very similar to the references and the best among the models created with a low PIDE template.
iTasser high PIDE models are not so accurate (RMSD=0.692). However, the low PIDE models are relatively good and come after Swiss-model (RMSD=1.232).
For Modeller: How does including more templates change the model quality?
Using 4 high PIDE templates did not improve much the results as we have expected. Probably combining the templates in one alignment may have introduced some "noise". In equivalent case of low PIDE templates, this noise became much worse and the RMDS has risen from 1.523 (1 low PIDE template) to 20.935 (3 low PIDE templates). However, adding one template with a high PIDE to a template with a low PIDE significantly improved the model quality (from 1.523 to 0.328).
Sources
Modeller:
1) N. Eswar, M. A. Marti-Renom, B. Webb, M. S. Madhusudhan, D. Eramian, M. Shen, U. Pieper, A. Sali. Comparative Protein Structure Modeling With MODELLER. Current Protocols in Bioinformatics, John Wiley & Sons, Inc., Supplement 15, 5.6.1-5.6.30, 2006.
2) M.A. Marti-Renom, A. Stuart, A. Fiser, R. Sánchez, F. Melo, A. Sali. Comparative protein structure modeling of genes and genomes. Annu. Rev. Biophys. Biomol. Struct. 29, 291-325, 2000.
3) A. Sali & T.L. Blundell. Comparative protein modelling by satisfaction of spatial restraints. J. Mol. Biol. 234, 779-815, 1993.
4) A. Fiser, R.K. Do, & A. Sali. Modeling of loops in protein structures, Protein Science 9. 1753-1773, 20005)
5) Tutorial [1]
Swiss-Model:
1) Arnold K., Bordoli L., Kopp J., and Schwede T. (2006). The SWISS-MODEL Workspace: A web-based environment for protein structure homology modeling. Bioinformatics, 22,195-201.
2) Schwede T, Kopp J, Guex N, and Peitsch MC (2003) SWISS-MODEL: an automated protein homology-modeling server. Nucleic Acids Research 31: 3381-3385.
3) Guex, N. and Peitsch, M. C. (1997) SWISS-MODEL and the Swiss-PdbViewer: An environment for comparative protein modeling. Electrophoresis 18: 2714-2723.
4) Benkert P, Biasini M, Schwede T. (2011). "Toward the estimation of the absolute quality of individual protein structure models." Bioinformatics, 27(3):343-50. (QMEAN-score)
5) Server [2]
iTasser:
1) Ambrish Roy, Alper Kucukural, Yang Zhang. I-TASSER: a unified platform for automated protein structure and function prediction. Nature Protocols, vol 5, 725-738 (2010).
2) Yang Zhang. I-TASSER server for protein 3D structure prediction. BMC Bioinformatics, 9:40 (2008).
3) Ambrish Roy, Jianyi Yang, Yang Zhang. COFACTOR: an accurate comparative algorithm for structure-based protein function annotation. Nucleic Acids Research, vol 40, W471-W477 (2012).
4) Server [3]
