A brief primer in energy minimization. Or, when you need it, how to do it and what can you expect.
Part I Global scores
Energy minimization is a mouthful. In the case of macromolecules it implies that any given experimentally determined structure is not at is energy minimum due to conformational strain, missing interactions, etc.
Note, I’ll be skipping tons of theory about modeling, force fields, etc. There are very good books and manuals that describe in detail all of those things. For example, check the gromacs website.
For very simple molecules energy minimization might be able to find the global minimum, that is, the conformation with the lowest energy possible. For large molecules we are bound to get stuck at a local minimum. Thus, the aim of energy minimization for proteins and other macromolecules is to relieve structural strain.
A beautiful example of what energy minimization can do is illustrated at the SPBDV viewer tutorial pages. In a nutshell, you can fix strained bonds and atoms positions but you won’t be able to generate hydrogen bonds or turn a helix into a sheet and so on. So, let’s address the questions put forth in the title, ok?
When do you need to perform energy minimization?
Well, knowing the limitations for what energy minimization (em) can do while acknowledging that experimentally determined structures are not perfect the answer should be apparent. You can probably reconstruct bad or even missing sidechains but not missing backbone. So, while em will improve (in general terms) the structure it will only be worth it if you are going to do something else with that structure. For example, using it for molecular dynamics, docking or maybe for homology modeling.
As to for the two other questions lets address them by actually performing em on an X-ray structure.
For this exercise I’ll use the 1YPI pdb and UCSF Chimera.
By running the structure through molprobity we get this results:
REMARK 40 ALL-ATOM CLASHSCORE : 34.57
REMARK 40 BAD ROTAMERS : 20.3% 81/400 (TARGET 0-1%)
REMARK 40 RAMACHANDRAN OUTLIERS : 1.0% 5/490 (TARGET 0.2%)
REMARK 40 RAMACHANDRAN FAVORED : 95.1% 466/490 (TARGET 98.0%)
After running the output from molprobity through a several “energy minimizers” these are the results:
Scwrl4*
REMARK 40 ALL-ATOM CLASHSCORE : 68.74
REMARK 40 BAD ROTAMERS : 0.3% 1/400 (TARGET 0-1%)
REMARK 40 RAMACHANDRAN OUTLIERS : 1.0% 5/490 (TARGET 0.2%)
REMARK 40 RAMACHANDRAN FAVORED : 95.1% 466/490 (TARGET 98.0%)
SPDBV (GROMOS96 force field)
REMARK 40 ALL-ATOM CLASHSCORE : 17.69
REMARK 40 BAD ROTAMERS : 2.2% 8/370 (TARGET 0-1%)
REMARK 40 RAMACHANDRAN OUTLIERS : 1.6% 8/490 (TARGET 0.2%)
REMARK 40 RAMACHANDRAN FAVORED : 91.4% 448/490 (TARGET 98.0%)
UCSF Chimera (Amber force field)
REMARK 40 ALL-ATOM CLASHSCORE : 1.46
REMARK 40 BAD ROTAMERS : 13.0% 52/400 (TARGET 0-1%)
REMARK 40 RAMACHANDRAN OUTLIERS : 0.8% 4/490 (TARGET 0.2%)
REMARK 40 RAMACHANDRAN FAVORED : 94.1% 461/490 (TARGET 98.0%)
YASARA server (Yasara force field)
REMARK 40 ALL-ATOM CLASHSCORE : 0.26
REMARK 40 BAD ROTAMERS : 2.5% 10/400 (TARGET 0-1%)
REMARK 40 RAMACHANDRAN OUTLIERS : 0.2% 1/490 (TARGET 0.2%)
REMARK 40 RAMACHANDRAN FAVORED : 96.1% 471/490 (TARGET 98.0%)
*Scwrl4 is not an energy minimization software per se. It is a sidechain conformation predictor.
Scwrl4 is the only program that makes the clashscore worse that the original. While it improves the bad rotamers it does not improve the Ramachandran favored residues. This is expected since Scwrl4 is not really designed for energy minimization after all. SPDBV viewer, a program that is oriented towards experimental structure determination, does a decent job for clashes and bad rotamers but the Ramachandran worsens. UCSF Chimera improves the clashscore, the bad rotamers (although not as good as SPDBV) with little impact on the Ramachandran.
I must admit that I was expecting UCSF Chimera to produce the best results or, at least, a good compromise between improvement of the clashes and worsening of the Ramachandran. As you can see from the data above, Yasara is the best way to minimize a structure. I was really surprised by this and, after reading the paper where the creation of the program is described (here and here), I believe I understand the results better. You see, Yasara was generated by adjusting/guiding the parametrization of the forcefield through Monte Carlo simulations without restraining the values to experimentally determined ones. They argue that they counterbalance the errors inherent to any forcefield. Their aim is to create a scheme where the energy minimization never worsens a X-Ray structure. It is an interesting read and their work seems to deliver.
Now, in contrast to the other software used here, the results from Yasara are produced in a proprietary format. However, the Yasara viewer is freely available upon registration. I guess my next visualization software review will be about that viewer.
To fulfill the objective of this post I want to emphasize that the numbers show above are global descriptors of the structures. They can give us a reference to judge a structure globally but won't allow us to check the quality of specific regions or residues. That will also be touch on a next blog post.
See you around.
- Posted using BlogPress from my iPad
Energy minimization is a mouthful. In the case of macromolecules it implies that any given experimentally determined structure is not at is energy minimum due to conformational strain, missing interactions, etc.
Note, I’ll be skipping tons of theory about modeling, force fields, etc. There are very good books and manuals that describe in detail all of those things. For example, check the gromacs website.
For very simple molecules energy minimization might be able to find the global minimum, that is, the conformation with the lowest energy possible. For large molecules we are bound to get stuck at a local minimum. Thus, the aim of energy minimization for proteins and other macromolecules is to relieve structural strain.
A beautiful example of what energy minimization can do is illustrated at the SPBDV viewer tutorial pages. In a nutshell, you can fix strained bonds and atoms positions but you won’t be able to generate hydrogen bonds or turn a helix into a sheet and so on. So, let’s address the questions put forth in the title, ok?
When do you need to perform energy minimization?
Well, knowing the limitations for what energy minimization (em) can do while acknowledging that experimentally determined structures are not perfect the answer should be apparent. You can probably reconstruct bad or even missing sidechains but not missing backbone. So, while em will improve (in general terms) the structure it will only be worth it if you are going to do something else with that structure. For example, using it for molecular dynamics, docking or maybe for homology modeling.
As to for the two other questions lets address them by actually performing em on an X-ray structure.
For this exercise I’ll use the 1YPI pdb and UCSF Chimera.
By running the structure through molprobity we get this results:
REMARK 40 ALL-ATOM CLASHSCORE : 34.57
REMARK 40 BAD ROTAMERS : 20.3% 81/400 (TARGET 0-1%)
REMARK 40 RAMACHANDRAN OUTLIERS : 1.0% 5/490 (TARGET 0.2%)
REMARK 40 RAMACHANDRAN FAVORED : 95.1% 466/490 (TARGET 98.0%)
After running the output from molprobity through a several “energy minimizers” these are the results:
Scwrl4*
REMARK 40 ALL-ATOM CLASHSCORE : 68.74
REMARK 40 BAD ROTAMERS : 0.3% 1/400 (TARGET 0-1%)
REMARK 40 RAMACHANDRAN OUTLIERS : 1.0% 5/490 (TARGET 0.2%)
REMARK 40 RAMACHANDRAN FAVORED : 95.1% 466/490 (TARGET 98.0%)
SPDBV (GROMOS96 force field)
REMARK 40 ALL-ATOM CLASHSCORE : 17.69
REMARK 40 BAD ROTAMERS : 2.2% 8/370 (TARGET 0-1%)
REMARK 40 RAMACHANDRAN OUTLIERS : 1.6% 8/490 (TARGET 0.2%)
REMARK 40 RAMACHANDRAN FAVORED : 91.4% 448/490 (TARGET 98.0%)
UCSF Chimera (Amber force field)
REMARK 40 ALL-ATOM CLASHSCORE : 1.46
REMARK 40 BAD ROTAMERS : 13.0% 52/400 (TARGET 0-1%)
REMARK 40 RAMACHANDRAN OUTLIERS : 0.8% 4/490 (TARGET 0.2%)
REMARK 40 RAMACHANDRAN FAVORED : 94.1% 461/490 (TARGET 98.0%)
YASARA server (Yasara force field)
REMARK 40 ALL-ATOM CLASHSCORE : 0.26
REMARK 40 BAD ROTAMERS : 2.5% 10/400 (TARGET 0-1%)
REMARK 40 RAMACHANDRAN OUTLIERS : 0.2% 1/490 (TARGET 0.2%)
REMARK 40 RAMACHANDRAN FAVORED : 96.1% 471/490 (TARGET 98.0%)
*Scwrl4 is not an energy minimization software per se. It is a sidechain conformation predictor.
Scwrl4 is the only program that makes the clashscore worse that the original. While it improves the bad rotamers it does not improve the Ramachandran favored residues. This is expected since Scwrl4 is not really designed for energy minimization after all. SPDBV viewer, a program that is oriented towards experimental structure determination, does a decent job for clashes and bad rotamers but the Ramachandran worsens. UCSF Chimera improves the clashscore, the bad rotamers (although not as good as SPDBV) with little impact on the Ramachandran.
I must admit that I was expecting UCSF Chimera to produce the best results or, at least, a good compromise between improvement of the clashes and worsening of the Ramachandran. As you can see from the data above, Yasara is the best way to minimize a structure. I was really surprised by this and, after reading the paper where the creation of the program is described (here and here), I believe I understand the results better. You see, Yasara was generated by adjusting/guiding the parametrization of the forcefield through Monte Carlo simulations without restraining the values to experimentally determined ones. They argue that they counterbalance the errors inherent to any forcefield. Their aim is to create a scheme where the energy minimization never worsens a X-Ray structure. It is an interesting read and their work seems to deliver.
Now, in contrast to the other software used here, the results from Yasara are produced in a proprietary format. However, the Yasara viewer is freely available upon registration. I guess my next visualization software review will be about that viewer.
To fulfill the objective of this post I want to emphasize that the numbers show above are global descriptors of the structures. They can give us a reference to judge a structure globally but won't allow us to check the quality of specific regions or residues. That will also be touch on a next blog post.
See you around.
- Posted using BlogPress from my iPad
Comments