I agree that it would certainly be bad if your model was inconsistent
with the Wilson B.  When I first ran into people who were worried about
this I recommended that they calculate the Wilson B of their model, not
look at the average B as a proxy.  You can calculate structure factors
from your model and feed those into a Wilson B calculator and get the
proper value.

   When you do this you find pretty quickly that a refinement program
has to be screwed up pretty badly for it to create a model with even a
small difference between the observed and calculated Wilson B.  If your
model is off only a little the R value shoots up!  Other than when
debugging refinement code, I have never seen a refined model where there
was any significant difference between the calculated and observed
Wilson B.  This problem simply doesn't come up.

   Try it for yourself and see.

   When you see a difference between the Wilson B and the average of the
B's in your model, all you are seeing is that you have a variety of B
factors and that is not a surprise.  I would be worried if all my B
factors were equal!

   High B factors are not a problem to be fixed!  Some atoms do move
more than others.  Why would you expect otherwise?

Dale Tronrud

On 10/14/2016 12:28 AM, Carlos CONTRERAS MARTEL wrote:
> Sunanda,
> As "common people", ... "agreement" don't means for me "equals" ...
> So I hope I'm not so "incorrect" if I keep one mind the Wilson B as the
> model
> refinement goes and the average B factor changes with different strategies.
> If you are still worry about your Bfactor, you could try TLS, using the
> right approach could
> improve your results.
> All the best
> Carlos
> On 10/13/16 18:42, Dale Tronrud wrote:
>>    It seems to be common for people to make the incorrect assumption
>> that the average of the atomic B factors in the PDB file should equal
>> the Wilson B.  The Wilson B is actually a weighted average of the
>> individual B factors where the weighting is rather mysterious, but the
>> small B factors have much higher weight than the larger ones. While the
>> Wilson B is simply a property of your data set the average of the atomic
>> B factors will also depend on choices you have made in building your
>> model.
>>    In your case the core of your protein does have B factors that match
>> your Wilson B.  The fact that you have loops that have larger B's
>> doesn't really mean there is a problem, because those atoms don't
>> contribute much to the higher resolution reflections and don't come into
>> the calculation of the Wilson B.
>>    The average of your B factors will increase if you choose to build
>> model for more disordered regions, while the Wilson B will, of course,
>> remain the same.  The average B factor will be larger, for example, if
>> you are more aggressive in building water molecules.  This does not
>> indicate a problem but is an unavoidable consequence of your choice to
>> take more risks.
>> Dale
>> On 10/13/2016 12:16 AM, sunanda williams wrote:
>>> Sorry for not making the problem clear!
>>> The overall B factor for the 3.0 A structure is around 98.00 A*2. Most
>>> of the deposited structures in the PDB site around this resolution
>>> has an av B around 70 A*2....
>>> All other statistics looks fine. I got a warning message while trying to
>>> upload the structure in PDB that 98 was higher than the norm!
>>> The reason why the structure has such a high B factor could be due to
>>> disordered loops... All the same I was wondering how acceptable this
>>> structure would be to picky reviewers..
>>> And 'better the B factors' means bring them down...sorry couldn't phrase
>>> it 'better'! I am going to try doing refinement with group B factors!
>>> Would that help?
>>> Thanks Robbie, I think this I need to make the best of the model I have.
>>> Prof. Sekar will try PDB_redo..Thanks!
>>> On Thu, Oct 13, 2016 at 11:41 AM, Pavel Afonine <pafon...@gmail.com
>>> <mailto:pafon...@gmail.com>> wrote:
>>>     I fully agree with Dale in not understanding what the problem is.
>>>     Perhaps I have a better chance if you clearly explain what exactly
>>>     you mean by "is there any way to better the B factors".
>>>     Pavel
>>>     On Thu, Oct 13, 2016 at 12:57 PM, Dale Tronrud
>>>     <de...@daletronrud.com <mailto:de...@daletronrud.com>> wrote:
>>>            I'm sorry but I don't understand what your problem is.  Do
>>>         you think
>>>         the B factors are too small for a 3A data set?  A range of 70 to
>>>         75 is a
>>>         little smaller than usual but probably not out of bounds.
>>>         Dale Tronrud
>>>         On 10/12/2016 7:59 PM, sunanda williams wrote:
>>>         > Hi all,
>>>         > I have a structure at 3.0 A and R/Rfree of 24/28. The mean B
>>>         value is
>>>         > around 98. The B value is especially high at the N terminus
>>>         and two loop
>>>         > regions (around 120-150 AA).
>>>         > The rest of the structure averaged around 70-75. Has anyone
>>>         faced such a
>>>         > scenario? How reliable is the structure and is there any way
>>>         to better
>>>         > the B factors.
>>>         > Any help is appreciated.
>>>         > Thank you!!

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to