Pavel has listed the papers that describe the underlying maths but unless
you have high resolution data (> 1.9A say?) unscrambling results from
occupancy plus B factor refinement is problematic to say the least..

Think about residues such as ARG or LYS where the termini are often
extremely wobbly - they are certainly in the crystal mix butt assigning a
single coordinate to those atoms is a bit of a lie..   However we do just
that thinking it is better to give some guidance to the community than to
assign OCC=0.0 .

It is a bit the same for your ligand problem.
I tend to start with both occs fixed to 0.5 and see what happens to be B
values - no ideal but it gives some feeling for the problem and you can
adjust the occupancies in the light of the first results. As Pavel says -
once you are reasonably close the an answer you can try the mathematical
approach..
Good luck Eleanor


On Thu, 23 Nov 2023 at 00:30, Pavel Afonine <[email protected]> wrote:

> Hi Matt,
>
> Im wondering about occupancy refinements - both what's going on under the
>> hood and what are best practices.
>>
>
> since you are quoting Phenix I suggest this bit of reading that is
> somewhat relevant to your questions:
>
> https://phenix-online.org/phenixwebsite_static/mainsite/files/newsletter/CCN_2015_07.pdf#page=12
> This documents 13 typical occupancy refinement scenarios and how they can
> be handled in Phenix.
>
>
>> In the example I have, there is a ligand found in two distinct, partially
>> overlapping sites that can be modeled is some confidence, but likely there
>> are very low occupancy additional poses that blurs the electron density.
>> The modeled poses are known from prior work, so even though there is
>> smearing we know the ligand is in the modeled conformation. After
>> perturbing the crystal these I am trying to decide what the best approach
>> is to get some sort of numerical occupancy value to describe the
>> distribution.
>>
>
> I apologize in advance for this trivial statement, but refinement +
> validation are the tools to answer your question.
> If you can model these poses with an atomic model and prove they match the
> experimental diffraction data ("The modeled poses are known from prior
> work" doesn't count in this regard), then you are all good! Depending on
> the size ration ligand:whole structure, R factors may or may not be useful
> quantifiers of the modeling choices. So your best bets may be local
> quantities, such as refined ligand group occupancy, flatness of Fo-Fc map
> (assessed as map values at atom positions), local map-model CC, etc. Try to
> challenge your modeling decisions by placing similar to expected answers
> but knowingly wrong models and see how that changes the fit/quality
> criteria -- that may give you a way to assess uncertainty.
>
> 1.  In Phenix, how is the occupancy number determined?
>
>
> Please refer to the paper I mentioned above, and let me know if you have
> additional questions.
>
>
>> Is there a real-space correlation between the experimental density and
>> the model(weighted to occ) that is optimized?
>
>
> Yes, but very indirectly through optimization of overall standard ML
> target function.
>
>
>>   How can this go wrong?
>
>
> It can go wrong in as many ways as the refinement target function has
> local minima, that is in many thousands ways!
>
>
>> I fear that the smearing and heterogenous nature will through the
>> refinement off (over or underfitting to periphery density rather than
>> hyper-localized position of the model)
>>
>
> That is a valid concern that is good to keep in mind but that is not a
> show-stopper!
>
>
>> 2.  Are there errors associated with the occupancy numbers?
>>
>
> Yes, like with any model refinable parameter. For some discussion please
> see:
>
> https://www.nature.com/articles/s41467-018-06957-w
>
>
>>
>> 3.  For my own testing, I did 5% increments and manually observed Fo-Fc
>> and 2Fo-Fc maps and selected a value that resulted in the lowest amount of
>> both positive and negative Fo-Fc peaks.  This is how we submitted the work
>> to journal but reviewer wants it to be automatically calculated.
>
>
> Above mentioned paper is now more relevant in light of this question! Yes,
> you can do manual sampling to get starting values (because the closer they
> are to the true values, the better chances refinement is successful), then
> do some refinement starting with these values.
>
> Good luck!
> Pavel
>
>
> ------------------------------
>
> To unsubscribe from the CCP4BB list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1
>

########################################################################

To unsubscribe from the CCP4BB list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1

This message was issued to members of www.jiscmail.ac.uk/CCP4BB, a mailing list 
hosted by www.jiscmail.ac.uk, terms & conditions are available at 
https://www.jiscmail.ac.uk/policyandsecurity/

Reply via email to