Re: [Chimera-users] fitting models into maps
Hi Dan, The Chimera fit model in map tool doesn't give a correlation value. To get a correlation requires you have two maps to compare, so first a map needs to be calculated from the PDB model that matches the resolution of the experimental map. Although I'd like to have that in Chimera, it is not currently there. I have used the EMAN pdb2mrc program to compute the map. Then you can use the Chimera fit map in map tool to get a correlation value, or you could use fit model in map, then open the map calculated for the PDB and find the correlation for the orientation obtained for the PDB model. You're right that reported correlation values depend on some often unreported parameters. First there is the method used for producing a map from the PDB model. Then the correlation depends on the part of the volume used. I would guess this is typically the portion within a certain contour level of the PDB map. What contour level? Something like 1 standard deviation is not sensible since it depends on the size of the map (the extent of padding). Probably a contour level that encloses a given volume -- maybe the volume enclosed by the solvent excluded surface of the PDB model with a reasonable probe radius. Would be nice if Chimera automated this calculation starting from the fit model in map position -- some day, hopefully in 2008. Reporting average density value at the atom positions would only make sense if the map is normalized. Maybe reporting that value in standard deviations. But the standard deviation depends on how much empty volume surrounds the structure, so that is not very satisfactory. Chimera reports that number so you can compare different fits in the same map -- only relative values are meaningful. Another value I have seen reported is the number of atoms outside a certain contour level of the experimental map. The fit model in map tool reports that in the reply log (Favorites / Reply Log). Of course it depends on how you set the contour level. Again if you set that as some number of standard deviations that depends on the map size. Alternatively you could set the contour level so the experimental map encloses the expected volume. Menu entry Tools / Volume Data / Measure Volume and Area can help set such a contour level. Seems correlation is the most meaningful number. You can calculate the standard deviation of a map from its mean in newer versions of Chimera with menu entry Tools / Volume Data / Volume Mean, SD, RMS. Tom Daniel Southworth wrote:
Tom:
Hi. I'm a post-doc in David Agard's lab and I have a question about the "fit model in map" function in chimera. I'm having some good success fitting crystal structures into my low-resolution em maps using this function once I get a global fit either by hand or using situs. I'm wondering if there is any way to get a correlation value out of the fits or how people typically report on their confidence of the chimera fitting in publications? I know correlation values are often different from program to program so perhaps the best measure is to state the average map value? Any help would be great - thanks.
Dan
Thanks - this is very informative. I agree it sounds like getting a cc value from 'fit map into map' is probably the best way. When I generate the volume from the pdb with eman I'll just have the molecular weight set to a threshold of 1 and use that threshold when I run 'fit map into map'. I'll check in to using a std value as well to see if the numbers are any different. Thanks for your help. Dan On Dec 19, 2007, at 9:50 PM, Tom Goddard wrote:
Hi Dan,
The Chimera fit model in map tool doesn't give a correlation value. To get a correlation requires you have two maps to compare, so first a map needs to be calculated from the PDB model that matches the resolution of the experimental map. Although I'd like to have that in Chimera, it is not currently there. I have used the EMAN pdb2mrc program to compute the map. Then you can use the Chimera fit map in map tool to get a correlation value, or you could use fit model in map, then open the map calculated for the PDB and find the correlation for the orientation obtained for the PDB model.
You're right that reported correlation values depend on some often unreported parameters. First there is the method used for producing a map from the PDB model. Then the correlation depends on the part of the volume used. I would guess this is typically the portion within a certain contour level of the PDB map. What contour level? Something like 1 standard deviation is not sensible since it depends on the size of the map (the extent of padding). Probably a contour level that encloses a given volume -- maybe the volume enclosed by the solvent excluded surface of the PDB model with a reasonable probe radius. Would be nice if Chimera automated this calculation starting from the fit model in map position -- some day, hopefully in 2008.
Reporting average density value at the atom positions would only make sense if the map is normalized. Maybe reporting that value in standard deviations. But the standard deviation depends on how much empty volume surrounds the structure, so that is not very satisfactory. Chimera reports that number so you can compare different fits in the same map -- only relative values are meaningful.
Another value I have seen reported is the number of atoms outside a certain contour level of the experimental map. The fit model in map tool reports that in the reply log (Favorites / Reply Log). Of course it depends on how you set the contour level. Again if you set that as some number of standard deviations that depends on the map size. Alternatively you could set the contour level so the experimental map encloses the expected volume. Menu entry Tools / Volume Data / Measure Volume and Area can help set such a contour level.
Seems correlation is the most meaningful number.
You can calculate the standard deviation of a map from its mean in newer versions of Chimera with menu entry Tools / Volume Data / Volume Mean, SD, RMS.
Tom
Daniel Southworth wrote:
Tom:
Hi. I'm a post-doc in David Agard's lab and I have a question about the "fit model in map" function in chimera. I'm having some good success fitting crystal structures into my low-resolution em maps using this function once I get a global fit either by hand or using situs. I'm wondering if there is any way to get a correlation value out of the fits or how people typically report on their confidence of the chimera fitting in publications? I know correlation values are often different from program to program so perhaps the best measure is to state the average map value? Any help would be great - thanks.
Dan
participants (2)
-
Daniel Southworth -
Tom Goddard