[Chimera-users] Fit model in map algorithm

[Tom Goddard]

Christopher Akey wrote:
> Tom-
>
> I was wondering how the fit model into map works? Is it calculating a 
> ccf and if so how since a pdb and a map file are quite different 
> entities.
>
>
> cheers  C Akey
Hi Christopher,

 The Chimera fit model in map capability does a rigid rotation and 
translation to maximize the sum of the map values at the atom 
positions.  It reports the average map value at the atom positions.  It 
doesn't report a correlation coefficient.  That would require 
calculating a simulated map from the PDB at the desired resolution and 
Chimera currently cannot do that.  I would like to add that capability 
since correlation values are the standard measure of goodness of fit 
that is reported in the literature.  The fit model in map code uses a 
gradient ascent iterative method to maximize the sum of map values.

 The Chimera fit map in map capability will report a correlation value.  
So you could run some external program to compute a simulated map (I've 
used pdb2mrc from the EMAN package) then fit that simulated map into 
your experimental map with the fit map in map dialog and get a 
correlation value.  Or you could use the Chimera fit model in map to fit 
the PDB, then open the simulated map, register it with the fit PDB 
(using Model Panel / Transform As button), and then have the fit map in 
map dialog report the correlation without optimizing the simulated map 
placement.  I'd expect  very similar results from fitting the PDB versus 
fitting the simulated map.

 The fit map in map algorithm maximizes the "overlap" between maps.  
That is not the same as the correlation.  It is a sum of the pointwise 
products between the two maps at grid points within the contour surface 
of one of the maps.  That differs from the correlation in that it is not 
normalized by the sum of the squares of each maps' values within the 
surface.  This has important consequences.  The Chimera "overlap" 
optimization will move the map into a nearby high density region even if 
the pattern of highs and lows in the density does not match.  The 
correlation normalization eliminates the tendency to move to higher 
density -- instead just optimizing the match of the highs and lows in 
the density.  Both are useful and I will in the future add an option to 
optimize the correlation.

 The Tools section of the Chimera User's Guide describes the map and 
model fitting methods:

http://www.cgl.ucsf.edu/chimera/docs/ContributedSoftware/fitmodels/fitmodels.html 

http://www.cgl.ucsf.edu/chimera/docs/ContributedSoftware/fitmaps/fitmaps.html 


   Tom