Re: [ccp4bb] Far to good r-factors
On Mon, May 31, 2010 at 9:15 PM, Dale Tronrud det...@uoxray.uoregon.edu wrote: One of the great mysteries of refinement is that a model created using high resolution data will fit a low resolution data set much better than a model created only using the low resolution data. It appears that there are many types of errors that degrade the fit to low resolution data that can only be identified and fixed by using the information from high resolution data. Is it such a mystery? Isn't it just a case of overfitting to the experimental errors in the low res data if you tried to use the same parameterization restraint weighting as for the high res refinement? Consequently you are forced to use fewer parameters and/or higher restraint weighting at low res which obviously is not going to give as good a fit. Cheers -- Ian
Re: [ccp4bb] Far to good r-factors
This would be a possible explanation, and certainly is a problem with low resolution refinements, but the free R indicates that overfitting is not the problem here. (I'm assuming that the proper choice of test set has been made in this case.) In my experience, for very isomorphous pairs of structures, when a high resolution model is used as the starting point for a low resolution refinement, even the R values before refinement will be very good and that means fitting the noise can't be the cause. Our methods today are simply not as good at fitting low resolution data in the absence of high resolution data as they are in its presence. Dale Tronrud On 06/01/10 04:51, Ian Tickle wrote: On Mon, May 31, 2010 at 9:15 PM, Dale Tronrud det...@uoxray.uoregon.edu wrote: One of the great mysteries of refinement is that a model created using high resolution data will fit a low resolution data set much better than a model created only using the low resolution data. It appears that there are many types of errors that degrade the fit to low resolution data that can only be identified and fixed by using the information from high resolution data. Is it such a mystery? Isn't it just a case of overfitting to the experimental errors in the low res data if you tried to use the same parameterization restraint weighting as for the high res refinement? Consequently you are forced to use fewer parameters and/or higher restraint weighting at low res which obviously is not going to give as good a fit. Cheers -- Ian
Re: [ccp4bb] Far to good r-factors
Paul, Does your lower resolution structure have the same unit cell as the model used for MR? If your two crystals are the same except for the presence of the ligand, then you need to make sure to keep the same Rfree set for both. Otherwise, some reflections that were previous in the Rwork set will now be in the Rfree set, but they are not really free because the model was previously refined using these reflections. Greg On May 30, 2010, at 8:15 AM, Paul Lindblom wrote: Hi everybody, once more I need your help. I solved the structure of an enzyme at resolution of 1.9 A. Now I was trying to get a complex and soaked some ligand to my crystals. I could solve the structure (and see poor density for my ligand or something else) at 3.0 A by molecular replacement using my 1.9A structure as a starting model. But the problem is now, that I got an R-work of 16.34 and an r-free of 20.23 for the new 3.0 A structure - without adding any waters or solvent/ligand molecules. The r-factors are even better than the ones I got for the 1.9A structure. So I think something is wrong with the whole thing. I observed twinning for both data and used the twin refinement option in refmac, but the results stay more or less the same. Any suggestions what to do? Thanks a lot, Paul -- Department of Biophysics Johns Hopkins University 302 Jenkins Hall 3400 N. Charles St. Baltimore, MD 21218 Phone: (410) 516-7850 (office) Phone: (410) 516-3476 (lab) Fax: (410) 516-4118 gdbow...@jhu.edu
Re: [ccp4bb] Far to good r-factors
This reminded me another thing... Did you create the free-R flags considering twinning? This is very important. Pavel. On 5/31/10 12:43 PM, Gregory Bowman wrote: Paul, Does your lower resolution structure have the same unit cell as the model used for MR? If your two crystals are the same except for the presence of the ligand, then you need to make sure to keep the same Rfree set for both. Otherwise, some reflections that were previous in the Rwork set will now be in the Rfree set, but they are not really free because the model was previously refined using these reflections. Greg On May 30, 2010, at 8:15 AM, Paul Lindblom wrote: Hi everybody, once more I need your help. I solved the structure of an enzyme at resolution of 1.9 A. Now I was trying to get a complex and soaked some ligand to my crystals. I could solve the structure (and see poor density for my ligand or something else) at 3.0 A by molecular replacement using my 1.9A structure as a starting model. But the problem is now, that I got an R-work of 16.34 and an r-free of 20.23 for the new 3.0 A structure - without adding any waters or solvent/ligand molecules. The r-factors are even better than the ones I got for the 1.9A structure. So I think something is wrong with the whole thing. I observed twinning for both data and used the twin refinement option in refmac, but the results stay more or less the same. Any suggestions what to do? Thanks a lot, Paul -- Department of Biophysics Johns Hopkins University 302 Jenkins Hall 3400 N. Charles St. Baltimore, MD 21218 Phone: (410) 516-7850 (office) Phone: (410) 516-3476 (lab) Fax: (410) 516-4118 gdbow...@jhu.edu mailto:gdbow...@jhu.edu
[ccp4bb] Far to good r-factors
Hi everybody, once more I need your help. I solved the structure of an enzyme at resolution of 1.9 A. Now I was trying to get a complex and soaked some ligand to my crystals. I could solve the structure (and see poor density for my ligand or something else) at 3.0 A by molecular replacement using my 1.9A structure as a starting model. But the problem is now, that I got an R-work of 16.34 and an r-free of 20.23 for the new 3.0 A structure - without adding any waters or solvent/ligand molecules. The r-factors are even better than the ones I got for the 1.9A structure. So I think something is wrong with the whole thing. I observed twinning for both data and used the twin refinement option in refmac, but the results stay more or less the same. Any suggestions what to do? Thanks a lot, Paul
Re: [ccp4bb] Far to good r-factors
Hi Paul, I've seen that type of behaviour before for low resolution structures. On such structures, either I have a very hard time getting at the same time a good geometry, good R-factors and satisfactory electron density, or things go very smoothly and all the statistics (model geometry, R-factors) plus electron density are fine. Too bad I have no way of predicting when things will be going well. Two examples where things went very smoothly: glyceraldehyde-phosphate dehydrogenase from Trypanosoma brucei brucei (PDB id: 2X0N re-refined fairly recently with Phenix); malate dehydrogenase from Archaeoglobus fulgidus (PDB id: 2X0I also re-refined with Phenix) The only thing you have to check is that the relative weighting of the X-ray term vs. the geometry term is appropriate, so that you do not lower the R-factors while the geometry is getting worse. HTH, Fred. Paul Lindblom wrote: Hi everybody, once more I need your help. I solved the structure of an enzyme at resolution of 1.9 A. Now I was trying to get a complex and soaked some ligand to my crystals. I could solve the structure (and see poor density for my ligand or something else) at 3.0 A by molecular replacement using my 1.9A structure as a starting model. But the problem is now, that I got an R-work of 16.34 and an r-free of 20.23 for the new 3.0 A structure - without adding any waters or solvent/ligand molecules. The r-factors are even better than the ones I got for the 1.9A structure. So I think something is wrong with the whole thing. I observed twinning for both data and used the twin refinement option in refmac, but the results stay more or less the same. Any suggestions what to do? Thanks a lot, Paul
Re: [ccp4bb] Far to good r-factors
Hi Paul, another hypothesis... If you take an ultra-high resolution structure from PDB (resolution higher than 1.0A), then cut the data at 3A and do some refinement, you will get unusually low R-factors. This may suggest that your crystal can diffract to higher resolution and 3A is not the limit. However, you mentioned twinning and I guess it would be wise to double check how the R-factor is computed in this case, and how the total model structure factor is defined. I recall some Garib's comments about the difference of R-factor values related to twinning... In addition, to make sure that what you observe is not refinement program dependent or how the R-factor and Fmodel are defined, I would try to re-compute the R-factors or do some quick refinement in another package. For example, does this command: phenix.model_vs_data model.pdb data.mtz give you the same R-factors? Good luck! Pavel. On 5/30/10 5:15 AM, Paul Lindblom wrote: Hi everybody, once more I need your help. I solved the structure of an enzyme at resolution of 1.9 A. Now I was trying to get a complex and soaked some ligand to my crystals. I could solve the structure (and see poor density for my ligand or something else) at 3.0 A by molecular replacement using my 1.9A structure as a starting model. But the problem is now, that I got an R-work of 16.34 and an r-free of 20.23 for the new 3.0 A structure - without adding any waters or solvent/ligand molecules. The r-factors are even better than the ones I got for the 1.9A structure. So I think something is wrong with the whole thing. I observed twinning for both data and used the twin refinement option in refmac, but the results stay more or less the same. Any suggestions what to do? Thanks a lot, Paul