Ad empirical observations:
I have run and inspected routinely literally hundreds of RSCC plots during
the progress of MR searches and autobuild/refinement cycles.
Almost always an initial average RSCC less than 0.6 to 0.55 indicated a
non-solution
or was otherwise unrecoverable. So I am perhaps a tad more pessimistic than
Eleanor
as far as MR maps go (the absence of bias in experimental maps may give the
slightly lower empirical CC threshold of 0.5).
The achievable CC of course is local, and the average depends, amongst
other factors mentioned, on the type of protein/crystal. Sturdy stuff like
helix bundles may well give average CCs of 0.95 or higher. Floppy
structures (without prejudice whether the plasticity is genuine or
reflects long-range order like packing issues) can be 0.85 without
real indications from the map what to improve.
In almost all cases, the average RSCC consistently reflected the
average fo vs fc correlation coefficients that Refmac reports.
In case of very large structures you may run into grid limitations,
which may also have some effect on the actual mean CC.
BR
-----Original Message-----
From: CCP4 bulletin board [mailto:[log in to unmask]] On Behalf Of
Eleanor Dodson
Sent: Friday, May 28, 2010 6:11 AM
To: [log in to unmask]
Subject: Re: [ccp4bb] How to calculate real-space CC by section?
I havent a reference for the "correct" value of the CC - it is just
based on maps I have seen solved then checked out later.
but if your final model gives a very poor CC with a map calculated from
experimental phases, either for parts of the structure, or for the
whole, it is time to worry. Maybe your phases are bad - poor
measurements, low solvent content, incorrect heavy atom sites, etc.
Or maybe your model has some serious errors?
But of course all crystal structures have some parts better ordered than
others and for those bits the exptlly phased map may have weak density,
espec. after solvent flattening.
eleanor
[log in to unmask] wrote:
> Hi Eleanor:
>
> Do you have some references in mind that discussed the value of CC (say
>> 0.5) to be able to build the structure? Didn't find one for right now:-(
>
> By the way, probably a "weak" question, In the case "a lousy model will
> give poor CCs even if the map is brilliant", we still accept this model
> dispite the poor CC, right? Sorry that I didn't get practically involved
> too much in real model building, but I just heard that model is more
> frequently built manually by eyes, not CC etc.
>
> Best Regards, Hailiang
>
>> If you ask for CORR SECTion then overlapmap does just that - the CC will
>> have a certain value for each section regardless of the CHAIN
>> parameters. If you want correlation residue by residue you must ask for
>> CORR RESI
>>
>> As someone said - a lousy model will give poor CCs even if the map is
>> brilliant..
>> But once your refinement is finished it is intresting to go back and
>> check the CC of the initial maps.
>>
>> There is a belief that you need a CC of >0.5 to be able to build the
>> structure but different problems and different builders achieve
>> different results..
>> Eleanor
>>
>> Hailiang Zhang wrote:
>>> Hi,
>>>
>>> I am working on a real space correlation on a specif protein section
>>> using
>>> CCP4 OVERLAPMAP. I am using the following scripts, not sure whether it
>>> is
>>> good or not (didn't find in OVERLAPMAP documentation).
>>>
>>> overlapmap \
>>> mapin1 ${PDB}-1.map \
>>> mapin2 ${PDB}-2.map \
>>> mapin3 ${PDB}-mask.map \
>>> <<eof
>>> CORR SECT
>>> CHAIN A $START $END
>>> END
>>>
>>> There is no error message, but the results make no difference no matter
>>> how I change $START and $END. I am not sure whether the above script is
>>> ok.
>>>
>>> By the way, more importantly to me, if corr sect works at all, will it
>>> print out a single CC value by integrating over the WHOLE region define
>>> by
>>> the section range?
>>>
>>> Thanks!
>>>
>>> Best Regards, Hailiang
>>
>>
>
>
|