Question about structure quality
Hi, I have a refined structure with phenix. The resolution=3.8 angstrom. Molprobity analysis shows following. Can anyone tell me if the structure is acceptable or I need to do something to improve it before going to publication? Thanks. Poor rotamers20.95%Goal: <1%Ramachandran outliers6.15%Goal: <0.2%Ramachandran favored80.13%Goal: >98%Cβ deviations >0.25Å23Goal: 0Residues with bad bonds: 0.00%Goal: 0%Residues with bad angles:0.75% -- ====================== Jason Structural Biology Department University of Pittsburgh ======================
On Tue, Apr 5, 2011 at 1:32 PM, Jason
I have a refined structure with phenix.
The resolution=3.8 angstrom. Molprobity analysis shows following. Can anyone tell me if the structure is acceptable or I need to do something to improve it before going to publication? Thanks.
Poor rotamers20.95%Goal: <1%Ramachandran outliers6.15%Goal: <0.2%Ramachandran favored 80.13%Goal: >98%Cβ deviations >0.25Å23Goal: 0Residues with bad bonds:0.00%Goal: 0% Residues with bad angles:0.75%
What's the clashscore? RMS(bonds), RMS(angles), R-work, R-free? The Ramachandran statistics are poor; I've seen worse published, but it would be wise to fix these. I'm assuming you don't have a high-resolution structure that you can use as a reference model - this is usually the best option. Otherwise, adding Ramachandran restraints will probably help a lot, but you should first fix all outliers manually in Coot (also applying real-space refinement with Coot's Ramachandran restraints turned on), as the default potential is very tight and can pull residues the wrong way if they're starting from a very bad position. The rotamer outliers are difficult to avoid without a reference model, unfortunately; I'd like to fix this, but it's going to take time. The c-beta deviations and bad angles should be fixed; this may require optimizing the X-ray/stereochemistry weight during refinement. However, they may result from sidechains sticking out of the density, which often results in local distortions. -Nat
Hi Jason,
The Ramachandran statistics are poor; I've seen worse published, but it would be wise to fix these. I'm assuming you don't have a high-resolution structure that you can use as a reference model - this is usually the best option. Otherwise, adding Ramachandran restraints will probably help a lot, but you should first fix all outliers manually in Coot (also applying real-space refinement with Coot's Ramachandran restraints turned on), as the default potential is very tight and can pull residues the wrong way if they're starting from a very bad position.
- Often using Ramachandran restraints fixes the problem right away, so I would probably do it first, and then walk through the list of outliers that you had before refinement run with Ramachandran restraints, and see *how* these outliers were fixed. Nat's suggestion should work too but might require more up-front work. - Run refinement with weights optimization (optimize_wxc=true); - Use NCS if available; - Secondary structure restraints should definitely help, but: -- you need to have secondary structure well defined in your input model if you want phenix.refine to pick it up automatically (and correctly), or alternatively -- define it manually in a parameter file and supply to phenix.refine. Pavel.
On Tue, Apr 5, 2011 at 1:57 PM, Pavel Afonine
- Often using Ramachandran restraints fixes the problem right away, so I would probably do it first, and then walk through the list of outliers that you had before refinement run with Ramachandran restraints, and see *how* these outliers were fixed. Nat's suggestion should work too but might require more up-front work.
It will "fix" the problem only to the extent that the Ramachandran statistics/plot look better - this does not automatically mean that the structure is more accurate. When you are refining against one of the primary validation criteria, more up-front work is essential. (The restraints will also not automatically fix every problem, so if you blindly turn them on without any rebuilding, the result will likely be worse than if you correct as many residues as possible yourself.) -Nat
After fixing the Ramachandran with restraints, I would recommend refining several cycles without restraints before depositing to see how many outliers stay in the allowed region. The ramachandran is often used as a measure of model quality, and you might just fool the end user into using your structure instead of a better model which has slightly worse ramachandran but much better than yours would be without restraints. eab Pavel Afonine wrote:
Hi Jason,
The Ramachandran statistics are poor; I've seen worse published, but it would be wise to fix these. I'm assuming you don't have a high-resolution structure that you can use as a reference model - this is usually the best option. Otherwise, adding Ramachandran restraints will probably help a lot, but you should first fix all outliers manually in Coot (also applying real-space refinement with Coot's Ramachandran restraints turned on), as the default potential is very tight and can pull residues the wrong way if they're starting from a very bad position.
- Often using Ramachandran restraints fixes the problem right away, so I would probably do it first, and then walk through the list of outliers that you had before refinement run with Ramachandran restraints, and see *how* these outliers were fixed. Nat's suggestion should work too but might require more up-front work.
- Run refinement with weights optimization (optimize_wxc=true);
- Use NCS if available;
- Secondary structure restraints should definitely help, but: -- you need to have secondary structure well defined in your input model if you want phenix.refine to pick it up automatically (and correctly), or alternatively -- define it manually in a parameter file and supply to phenix.refine.
Pavel.
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
I agree with Ed - it's a matter of balance! Also, it's good to realize that at low resolution even if your starting model has zero Ramachandran outliers it will most likely has a lot of outliers after a quick refinement without Ramachandran restrains - I've done this test for all low resolution structures in PDB. This is simply because the density at that resolution is not informative enough to keep the residues within the allowed Ramachgandran plot areas, and the other geometry restraints do not have that information either. So I guess at that low resolution the Ramachandran plot is rather the tool to get your model physically correct, rather than a validation tool (simply because you don't have much choice). Of course, keeping eye on the map and model never hurts, especially in this case. Pavel. On 4/5/11 2:11 PM, Edward A. Berry wrote:
After fixing the Ramachandran with restraints, I would recommend refining several cycles without restraints before depositing to see how many outliers stay in the allowed region. The ramachandran is often used as a measure of model quality, and you might just fool the end user into using your structure instead of a better model which has slightly worse ramachandran but much better than yours would be without restraints.
eab
Pavel Afonine wrote:
Hi Jason,
The Ramachandran statistics are poor; I've seen worse published, but it would be wise to fix these. I'm assuming you don't have a high-resolution structure that you can use as a reference model - this is usually the best option. Otherwise, adding Ramachandran restraints will probably help a lot, but you should first fix all outliers manually in Coot (also applying real-space refinement with Coot's Ramachandran restraints turned on), as the default potential is very tight and can pull residues the wrong way if they're starting from a very bad position.
- Often using Ramachandran restraints fixes the problem right away, so I would probably do it first, and then walk through the list of outliers that you had before refinement run with Ramachandran restraints, and see *how* these outliers were fixed. Nat's suggestion should work too but might require more up-front work.
- Run refinement with weights optimization (optimize_wxc=true);
- Use NCS if available;
- Secondary structure restraints should definitely help, but: -- you need to have secondary structure well defined in your input model if you want phenix.refine to pick it up automatically (and correctly), or alternatively -- define it manually in a parameter file and supply to phenix.refine.
Pavel.
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
Thanks everyone, ramachandran restrain does bring the ramachandran favored
to 91%. However, it does does improve the rotamer outlier (even a little
worse). After manually adjust the outliers in coot, which took me ~4hours
for one round, the rotamer outlier decreases to ~9%, which is much better. I
will keep working to see if there is more room for such improvement.
On Tue, Apr 5, 2011 at 5:22 PM, Pavel Afonine
I agree with Ed - it's a matter of balance!
Also, it's good to realize that at low resolution even if your starting model has zero Ramachandran outliers it will most likely has a lot of outliers after a quick refinement without Ramachandran restrains - I've done this test for all low resolution structures in PDB. This is simply because the density at that resolution is not informative enough to keep the residues within the allowed Ramachgandran plot areas, and the other geometry restraints do not have that information either. So I guess at that low resolution the Ramachandran plot is rather the tool to get your model physically correct, rather than a validation tool (simply because you don't have much choice). Of course, keeping eye on the map and model never hurts, especially in this case.
Pavel.
On 4/5/11 2:11 PM, Edward A. Berry wrote:
After fixing the Ramachandran with restraints, I would recommend refining several cycles without restraints before depositing to see how many outliers stay in the allowed region. The ramachandran is often used as a measure of model quality, and you might just fool the end user into using your structure instead of a better model which has slightly worse ramachandran but much better than yours would be without restraints.
eab
Pavel Afonine wrote:
Hi Jason,
The Ramachandran statistics are poor; I've seen worse published, but
it would be wise to fix these. I'm assuming you don't have a high-resolution structure that you can use as a reference model - this is usually the best option. Otherwise, adding Ramachandran restraints will probably help a lot, but you should first fix all outliers manually in Coot (also applying real-space refinement with Coot's Ramachandran restraints turned on), as the default potential is very tight and can pull residues the wrong way if they're starting from a very bad position.
- Often using Ramachandran restraints fixes the problem right away, so I would probably do it first, and then walk through the list of outliers that you had before refinement run with Ramachandran restraints, and see *how* these outliers were fixed. Nat's suggestion should work too but might require more up-front work.
- Run refinement with weights optimization (optimize_wxc=true);
- Use NCS if available;
- Secondary structure restraints should definitely help, but: -- you need to have secondary structure well defined in your input model if you want phenix.refine to pick it up automatically (and correctly), or alternatively -- define it manually in a parameter file and supply to phenix.refine.
Pavel.
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
-- ====================== Jason Structural Biology Department University of Pittsburgh ======================
participants (4)
-
Edward A. Berry -
Jason -
Nathaniel Echols -
Pavel Afonine