[phenixbb] DEN Refinement in phenix 1.7.3

[Jeff Headd]

I also meant to point out that if you are working at moderate
resolution (say 2.5 to 3.2A) and your goal is to use a reference model
to improve the quality of your refined model, it will be much faster
to use reference model restraints in phenix.refine. DEN is very useful
in allowing for large domain shifts that may be differences between
your working model and reference model, but if you know that they
should be similar, reference model restraints may be more useful to
you.

Jeff

On Tue, Jan 24, 2012 at 9:34 AM, Jeff Headd <jjheadd at lbl.gov> wrote:
> Hi Eike,
>
> Yes, DEN restraints can be activated in phenix.refine from the
> command-line with the current version and latest nightly builds, and
> they are the same deformable elastic network restraints available in
> CNS. We have been working closely with Axel Brunger and Gunnar
> Schroder to implement DEN in Phenix.
>
> We have not yet officially announced the DEN restraints as they are
> still being tested and actively developed to get the implementation
> just right, and the parameterization is still very much in flux. I
> hope that by the next version it will become and a stable feature, and
> at that point DEN will be added as an option in the GUI.
>
> These restraints have been shown to be particularly useful at low
> resolution, and there has been success in using at 4-5A and below. I'm
> not sure how useful they would be at relatively high resolution (say
> 2.5A or higher), as there are other restraint methods that work well
> at that resolution range that are far less computationally intensive.
>
> In almost all cases it is best to optimize the gamma and weight
> parameters, which is quite time intensive but is most likely to give
> you the best results. Currently this can be parallelized, but only on
> cores that share memory. I'm working on an implementation that will be
> compatible with SGE, but that is not yet ready. If you do optimize the
> gamma and weight parameters, you cannot simultaneously optimize B
> factor weights, which is another limitation that will be overcome in
> the future.
>
> As soon as a stable version is announced in the context of a new
> release, documentation will be available.
>
> To use them with the current release (1.7.3), you can use a
> parameterization such as this:
>
> refinement {
>  main {
>   den_refinement = True
>   number_of_macro_cycles = 1
>   nproc = 8
>  }
>  refine {
>   strategy = *individual_sites individual_sites_real_space rigid_body \
>              *individual_adp group_adp tls occupancies group_anomalous
>  }
>  den {
>   reference_file = reference.pdb
>   optimize = True
>   annealing_type = *torsion cartesian
>   final_refinement_cycle = True
>  }
> }
>
> If you have specific questions on how to run a refinement, please let me know.
>
> Thanks,
> Jeff
>
>
> On Tue, Jan 24, 2012 at 5:19 AM, Eike Schulz <eike.schulz at embl.de> wrote:
>> Dear phenixbb,
>>
>> I have noticed that in (since?) the latest release it is possible to
>> activate DEN refinement in phenix.refine. I assume this refers to deformable
>> elastic networks – right? What is your experience with DEN refinements ? As
>> I could not find any reference to it in the documentation I would be happy
>> to have feedback on when to use and when to avoid it e.g. resolution limits
>> etc. and what to watch out for.
>>
>> Many thanks in advance
>>
>> Eike
>>
>>
>>
>> _______________________________________________
>> phenixbb mailing list
>> phenixbb at phenix-online.org
>> http://phenix-online.org/mailman/listinfo/phenixbb
>>