[Chimera-users] CC calculation

[Tom Goddard]

Hi Qing,

   As you observed, different programs report different correlation 
values.  You can get any correlation value you want by changing the 
region over which the correlation is computed, or the formula for the 
correlation, or the resolution of the simulated map.

   First Chimera 1.5 or newer reports two correlation values: the 
correlation about the mean and the correlation about 0.   It is typical 
for EM maps to report the correlation about zero which is usually much 
higher.  If you use the old Chimera 1.4 it only reports the lower 
correlation about mean.  Formulas for these two correlation values are 
given in the Chimera manual.

   Next you need to choose the region over which the correlation is 
calculated.  Your approach of setting the simulated map contour level so 
the enclosed volume is what is expected is good.  Make sure you list the 
simulated map first in the measure correlation command since the first 
volume determines the region for the computation.

   Lastly, the simulated map resolution should match the map 
resolution.  There are different definitions of resolution used by the 
EM community -- it is controversial.  In any case you want the simulated 
map to match the level of detail seen in the experimental map.  If it 
doesn't match then you will get lower correlation.   You may need to try 
a range of resolutions.  The molmap documentation describes the 
different options for how the resolution value is used to make the 
simulated map.

   Unfortunately reported correlation values in publications rarely 
describe all the above choices so the values are not very meaningful or 
reproducible.

     Tom

-------- Original Message --------
Subject: Re: [Chimera-users] CC calculation
From: Qing Xie
To: Tom Goddard
Date: 1/26/12 11:31 AM
> Hi, Tom,
> Thanks for your help!
> I'm helping to model a structure (.pdb) into a EM density map.... This 
> is my first time using Chimera. I got a CC of ~0.87 after refining the 
> structure against the EM map using Rsref in CNS. I was suggested to 
> cross-check the CC by using Chimera since Chimera is more popular in 
> EM community.
> I calculated the simulated map following your email, but the CC varies 
> a lot by choosing different contour level for the EM map (#0). The EM 
> map's range is from -0.12 to 0.17 (sigma 0.013), while the simulated 
> map's range is from 0 to 2.64 (sigma 0.20). After I tried to claculate 
> the volumes and then matchs the volume of (#0) to (#1), CC is only ~0.63.
> I also tried CCP4 overlapmap and it gave me a CC of ~0.73.
> Do you have any other suggestions?
> Thank you very much!
> Qing
>



Tom Goddard wrote:
> If the PDB (#1) is already fit to the map (#0) you could create a 
> simulated map for the pdb then measure its correlation with the 
> experimental map:
>
>     molmap #1 12.5
>     measure correlation #2 #1
>
> The 12.5 in the molmap command is the resolution for the simulated map 
> which should match the experimental map.  By default the correlation 
> will be measured within the displayed contour level of the first map 
> given in the command.
>
> The commands are described in the Chimera User's Guide.
>
> http://www.cgl.ucsf.edu/chimera/current/docs/UsersGuide/midas/molmap.html
> http://www.cgl.ucsf.edu/chimera/current/docs/UsersGuide/midas/measure.html#correlation
>
>    Tom
>
>
>
>
>> Hi,
>> Can anyone help me with calculating correlation coefficient from a 
>> .pdb model to a .mrc map?
>> Thank you!
>> Qing
>>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://plato.cgl.ucsf.edu/pipermail/chimera-users/attachments/20120126/4b9215c0/attachment.html>