We generated this model in fully-automated mode, but there are several ways to get some idea as to how well the process went, and therefore how good we expect the model to be. In the SwissModel results page, scroll down to the section headed Alignment.
This section reports the sequence alignment that was used between the target sequence you wish to model and the sequence of the template structure. This is the first place to look if you want to get an idea how good the model is likely to be.
Notice also that the alignment shows secondary-structure assignments (h for 'helix', e for 'extended' or β-strand and a blank for 'coil') beneath the alignment.
Look for any differences between the secondary structure assignments from the template structure and predictions for the target sequence.
Differences might imply problems, but remember that, for the target, it is only a prediction and that even assignment from a known structure can be difficult.
Study the alignment carefully. Do you notice anything that concerns you? Think about what is likely to be difficult to model accurately - especially regions where there is no information from a template structure and where such regions occur.
Once you've had a think click here for some discussion.
As discussed earlier, we cannot know whether it is really correct until we have an actual structure of the protein in question. However, it is useful to assess the model to see whether it is likely to be correct.
In addition to the items discussed earlier that influence the quality of a model (primarily the sequence identity with the template and the quality of the resulting alignment), we can assess whether it is likely to be an energetically favourable and realistic conformation. Two popular programs for doing this are ANOLEA and MODCHECK.
The SwissModel server automatically provides the ANOLEA service for checking how good models are. It uses a 'knowledge-based' approach to verifying how realistic models are. Essentially it contains information about the preferred spatial relationships between residue types in real protein structures. These are used statistically to assess the similarity of the spatial distribution of residues in the model to that observed in real proteins. A score can then be provided for how much the model 'looks' like a real protein.
Of course we could calculate the actual energy of a protein as well. For any pair of bonded atoms, we know the optimum bond length: We can treat the bond as a spring obeying Hooke's Law, measure the observed bond length in the model and, from the deviation from the optimum, we can calculate an energy. Similarly, for triplets of atoms we know the optimium bond angle which can also be treated as a spring. We can also consider torsion (or twist) angles, van der Waals packing energy, electrostatics and hydrogen bond energy. A force field calculated from these data is often used in refining the model to fix bad geometry and clashes using molecular dynamics or energy minimisation. The main problem however, is that these molecular mechanics approaches are not able to account for the hydrophobic effect: the tendency of hydrophobic sidechains to pack away from the solvent for entropic reasons.
Because molecular mechanics potentials cannot account properly for the hydrophobic effect, knowledge-based, statistical 'pseudo-potentials' are a better way of evaluating whether a protein model is likely to be correct since they are based on analysis of real, high-quality protein structures and implicitly account for things like the hydrophobic effect.
Scroll down to the section titled Anolea / Gromos / Verify3D in the SwissModel results page. Other parameters are also calculated, but for the purposes of this practical, those results have been disabled.
The graph shows regions in green and red - the green regions are likely to be correct while there may be problems in the red areas.
The usual way of addressing these 'problem' areas is to examine the alignment carefully in these regions, hand-modify the alignment and repeat the model building process.
Another tool often used to evaluate models is 'MODCHECK'. Again, rather than use this online, we will look at pre-calculated results.
MODCHECK is part of package called 'WHATIF'. Most of the programs are aimed at checking PDB files generated by crystallographers, and therefore include checks for correct naming of atoms and residues etc. However, there are several indicators of model quality which apply equally to experimental and theoretical protein structures and MODCHECK is a subset of these tools.
Results from a run of MODCHECK are available here.
There are three ways of seeing the results, which are reproduced here primarily for authenticity. The list of results by 'one method at a time' is the most useful however, as it provides a brief description of the method (as well as a link to more detailed information) and what the information presented in the page actually means (which is somewhat of an omission from the tables) .
These results provide an in-depth look at the quality of the model in particular regions giving, for example, the degree to which bond lengths and angles are realistic.
Have a look at the results from the BNDCHK, ANGCHK and CHICHK methods. These check the bond lengths, bond angles and torsion angles respectively, and see if there are any unusual values.