Further assessment of homology model quality

The tools available in SwissPDB Viewer may also be used for quick model evaluation. The program uses mean force potential calculations of the energy of each residue in the model and displays the results in the form of a graph. To do this we need to load the optimized homology model file into SPDBV.
You may open the homology model file and the experimental 1g8p to see if there are any differences. The easiest will be to display the structures as Cα-alpha trace and color them differently. You will immediately notice the two regions where the structures differ from each other.
• For energy calculations make sure the current layer is the
model, and click on the little white arrow located at the right of the help question mark of the Align Window. The window expands, and displays a curve showing the energy of each residue (interactions with surrounding atoms). If there are no bad contacts, the energy is around or below zero, whereas bad contacts will have high energy above the zero line (red regions). You may also chose "Color: Force filed", which will color the model and show regions with high energy.
However, the energy analysis provided by the modeling server is probably more comprehensive.

At the early stages of homology modeling you can also evaluate how good your model is by using the "select aa making clashes" items of the "Select" menu. This will allow you to quickly focus on potentially problematic regions (holding the option key while you select these will not only select aa but also draw the clashes in pink on the screen). You can then choose the "Fix Selected Side chains (quick and dirty)" item of the "Tools" menu, which will browse the rotamer library to choose the best rotamer (the same commands are used if you want to replace an amino acid by another). By repeating the "Select aa making clashes" process, you should see that far less amino-acids are making problems. If not, this is probably a good clue that your threading (by other words the sequence alignment) is incorrect.
Important Note: Fixing side chains is just for you to evaluate the preliminary model prior to submitting it to the server. It will have little influence on model building and the quality of the final model, as the server reconstructs side chains during that process.

Other criteria for the quality of the homology model include model geometry and particularly the
Ramachandran plot. The Ramachandran plot may be checked in DeepView, you need to go to the Wind menu of the program and choose "Ramachandran", then in the "Select" menu choose "ALL". This will display all the torsion angles of the model in the Ramachandran plot. Pointing at any point in the plot will show the residue name. This way we may be able to check if there are any residues with bad torsion angles. A more comprehensive way of checking the geometry of the model is to use one of the dedicated servers. A list of internet-based servers for model validation can be found here, otherwise you may just jump to the JCSG Protein structure validation server (Joint Center for Structural Genomics) and submit your model for evaluation using programs like Procheck:

My Image

After submitting you will get en e-mail with a link to the structure validation. My output looked like this:

My Image

By clicking on any of the links above I get a detailed description of the corresponding parameter. For example, the Ramachandran plot:

My Image

And also detailed analysis showing the torsion angles for all amino acids in the protein (one aa type a time, total of 20), like here for Val and Tyr:

My Image

The analysis indicates that there are 17 residues with bad torsion angles. This is not surprising, we got an indication that this was the case when we looked at the energy in the previous page. One probably needs to go back and check these residues carefully and try to understand the problem. Another possibility is to have a look a the Procheck output for the modeling template, in this case 1g8p. To do that we don't need to submit the coordinates to the verification server, we can just go to PDBsum, find 1g8p and click the Ramachandran plot symbol on the right, and on the page which will appear click "Procheck" next to generate Procheck Analysis:

My Image

The analysis report will show that essentially the same residues in the 1g8p structure have problems with their Ramachandran angles. By other words, the homology model has just inherited the problems of the template! We could also be more curious and check the electron density for the experimental structure to see if it has any problems within the regions where we get bad torsion angles. To do that we need to use the electron-density server, EDS. While there, enter the PDB code, which brings us to the following page:

My Image

Down in this page we can start the Astex viewer, which will display the electron density of the molecule. We may easily center on the residues we are interested in by clicking on the sequence below the graphics window (not shown here). There are several options in the program which may modify the view of the model and the electron density. Below I am showing on the left one of the regions which has good quality electron density and on the right one of the regions, which had bad torsion angles, and apparently weak electron density:

My Image
My Image

Apparently, the weak density has contributed to bad geometry in this region of the structure. This example shows how important the quality of the electron density is for model quality. It also shows that problems present in the template are imported to the homology model.