[phenixbb] Weighting in phenix.real_space_refine

[Oliver Clarke]

Thanks Pavel!

So just to be sure - the average weight obtained with the randomly selected fragments is used for the entire model? Or different weights are used for different regions? I guess the latter would be in some ways preferable for large structures in cryoEM where there is a substantial variation in local resolution?

Cheers,
Oli
> On Sep 13, 2016, at 10:54 AM, Pavel Afonine <pafonine at lbl.gov> wrote:
> 
> Hi Oliver,
> 
> unlike reciprocal space refinement where optimization of the weight requires systematically trying full refinement (which is very slow unless you use many cpus), in real space this can be done very quickly and the speed does not depend on model or map size.
> The weight calculation procedure implemented in phenix.real_space_refine includes splitting the model into ten randomly picked continues ten-residue long segments, and finding the best weight for each segment. The best weight is considered to be the one that results in a model possessing pre-defined bond and angle rms deviations and that has best model-to-map fit among all trial weights. The obtained array of weights is filtered for outliers and the average weight is calculated and used as the best weight.
> 
> Parameters defining bond/angle rmsds as targets for weight optimization:
> 
>    target_bonds_rmsd = 0.01
>    target_angles_rmsd = 1.0
> 
> Pavel
> 
> On 9/12/16 14:14, Oliver Clarke wrote:
>> Hi all,
>> 
>> How does phenix.real_space_refine deal with restraint weighting? Looking at the output of the program, I get the impression that it splits the model up into  different chunks, and alters weights locally based on some model/map correlation metric - is this the case (and are there more specific details available), or how does it work otherwise?
>> 
>> Cheers,
>> Oliver.
>