[phenixbb] TLS, B factors, phenix and refmac

[Pavel Afonine]

I think what you describe below is a bit of re-inventing the wheel (in 
some sense, not completely). Here is why:

phenix.refine has an extremely complex algorithm of refinement ADP. By 
refining ADP I mean refining of all U=Utls+Ucryst+Uresidual. Briefly:
- it does some group iso B refinement to get starting TLS values;
- then it "simultaneously" refines TLS parameters and residual B;
- then it extracts TLS components from total B as described in 
http://www.ccp4.ac.uk/newsletters/newsletter45.pdf;
- it monitors to make sure that all parameters are meaningful at all times;
- then it repeats the whole process at next macro-cycle.

Look TLS related code in phenix.refine for more details.

The all details and parameters of the above algorithm were highly 
optimized using systematic re-refinement of 355 models selected from 
PDB. This makes ADP refinement (TLS+B+etc) in phenix very stable. See 
dedicated slide here, for actual results:
http://phenix-online.org/download/documentation/cci_apps/phenix_refine_quick_facts.pdf

At some point, I re-refined all models in PDB (that have data) using TLS 
refinement option in phenix.refine. It never crashed or got "unstabale". 
So, I don't think there is anything to improve in terms of stability of 
TLS refinement in PHENIX.

Please let me know if you find a case where this algorithm implemented 
in phenix.refine fails and I will try to fix it asap.

Cheers,
Pavel.



On 4/1/2008 1:01 PM, Mischa Machius wrote:
> Hi - Prompted by the recent discussions on B values, TLS refinement  
> and differences between Phenix and refmac, we looked into these  
> matters in more detail. We found that the crux of the problem lies in  
> the fact that TLS and B value refinements are usually decoupled. We  
> have developed a formalism that rolls both TLS and B value refinement  
> into one. Phenix and refmac were modified to carry out the  
> calculations, and the outputs from both programs were made compatible  
> to allow proper comparison of the results.
>
> We found that the stability of the refinements is now vastly improved.  
> More importantly, however, due to the reduced number of parameters,  
> these calculations can be carried out to resolutions of 7 Å with  
> meaningful representations of indiviual, anisotropic atomic  
> displacement parameters. This low-resolution limit required  
> reformulating the calculation of Wilson B values, but that is only a  
> minor aspect of our treatment that can be neglected.
>
> The new, combined procedure for the simultaneous refinement of TLS/B  
> is called 'TBS' refinement, reflecting all required components:  
> Translation, Bibation, Screw.
>
> Interestingly, the ‘T’ component is fairly insensitive to input  
> parameters, whereas the overall quality of the refinement is greatly  
> dependent on the ‘B’ component. The more emphasis is put on ‘B’, the  
> more convincing the results. There is a limit, though. At very high  
> levels of ‘B’, the so-called ‘bibacity limit’, the refinement becomes  
> very unstable, leading to inversion in severe cases. Seasoned  
> crystallographers familiar with the concepts can successfully push the  
> procedure to quite high 'B limits', whereas less experienced  
> practitioners should follow the protocols very carefully.
>
> Please contact us for any details.
>
> Best - MM
>
>
> --------------------------------------------------------------------------------
> Mischa Machius, PhD
> Associate Professor
> UT Southwestern Medical Center at Dallas
> 5323 Harry Hines Blvd.; ND10.214A
> Dallas, TX 75390-8816; U.S.A.
> Tel: +1 214 645 6381
> Fax: +1 214 645 6353
>
>
>
> _______________________________________________
> phenixbb mailing list
> phenixbb at phenix-online.org
> http://www.phenix-online.org/mailman/listinfo/phenixbb
>