[phenixbb] refinement

Thu Dec 10 21:56:43 PST 2009

Hi, 

Thanks for answering my previous question about weight adjustment, but I have a new problem.

If I use,

ordered_solvent=true simulated_annealing=true \

strategy=rigid_body+individual_sites+individual_adp \

main.number_of_macro_cycles=10 \

output.prefix=best_model --overwrite

the program run fine and give the result: Start R-work = 0.4713, R-free = 0.4786, Final R-work = 0.2721, R-free = 0.3769

However, when I tried to change weight,

ordered_solvent=true simulated_annealing=true \

strategy=rigid_body+individual_sites+individual_adp \

optimize_wxu=true optimize_wxc=true \

main.number_of_macro_cycles=10 \

output.prefix=best_model --overwrite

the program show error message, 

---------xray term weight optimization (wxc)----------            

Traceback (most recent call last):

  File "/usr/local/phenix-1.3b-rc6/phenix/phenix/command_line/refine.py", line 5, in <module>

    command_line.run(command_name="phenix.refine", args=sys.argv[1:])

  File "/usr/local/phenix-1.3b-rc6/phenix/phenix/refinement/command_line.py", line 82, in run

    call_back_after_monitor_collect=call_back_after_monitor_collect)

  File "/usr/local/phenix-1.3b-rc6/phenix/phenix/refinement/driver.py", line 1143, in run

    call_back_after_monitor_collect = call_back_after_monitor_collect)

  File "/usr/local/phenix-1.3b-rc6/phenix/phenix/refinement/strategies.py", line 449, in refinement_machine

    log                           = log)

  File "/usr/local/phenix-1.3b-rc6/phenix/phenix/refinement/weight_xray_chem.py", line 256, in __init__

    self.wxc_optimization()

  File "/usr/local/phenix-1.3b-rc6/phenix/phenix/refinement/weight_xray_chem.py", line 511, in wxc_optimization

    wxc = self.holder.wxc

AttributeError: weight instance has no attribute 'holder'

How do I fix this?

PC

Date: Thu, 10 Dec 2009 08:31:25 -0800
From: PAfonine at lbl.gov
To: phenixbb at phenix-online.org
Subject: Re: [phenixbb] refinement

Hi Pei-Chun,

the target function for coordinates refinement that is used in
phenix.refine is following:

T = wxc_scale * wxc * Txray + wc * Trestraints

where: wxc is determined automatically, wc =1 by default, and wxc_scale
= 0.5 by default. A user can adjust wxc_scale or wc. 

Responding your question about "put much more weight on geometry
restraints": you can do it either by decreasing wxc_scale or by
increasing wc. I would try an array of values for wxc_scale to see
which one gives the best result. 

Please let me know if it's still not clear or if you have any other
questions!

Pavel.

On 12/10/09 7:38 AM, Pei-Chun Lin wrote:

  Hi,

When talking about "put much more weight on geometry
restraints to avoid overfitting", what wxc and wxu value will you
suggest to use for a 4A structure?

I am working on a 3.5A structure and have no experience of adjusting
weight value before. Any suggestion will be very helpful to me.

PC

  Date: Tue, 8 Dec 2009 20:12:29 -0800

From: PAfonine at lbl.gov

To: phenixbb at phenix-online.org

Subject: Re: [phenixbb] refinement

Hi,

although I've seen cases like yours when SA in Cartesian space did
miracles, in many of cases modeling at 4A resolution is a tough
problem. 

Since clearly SA didn't work out (given R/Rfree ~ 37/50% and the
starting 48/50), I would try splitting your model into rigid domains
and refine each as a rigid body. Plus, I would refine one isotropic B
per residue or per domain (try both). This is a few minutes to try so
why don't you do so.

Also, I would still combine it (rigid body and B-factor refinement)
with Cartesian SA, but I would put much more weight on geometry
restraints to avoid overfitting.

As you probably aware of, in phenix.refine you can run combined
refinement jobs, for example, consisting of SA, rigid body, B-factor
refinement and so on. So I would recommend playing with the above
suggestions to see if it works. Plus, don't forget looking at the
maps - may be you manage to fit something manually.

By the way, which R-factors you get if you run:

phenix.model_vs_data data.mtz model.pdb

(phenix.model_vs_data will account for twinning (if any) using Xtriage
automatically) ?

Write us if you have questions.

Good luck,

Pavel.

On 12/8/09 7:02 PM, r n wrote:

    Thanks
a
lot. I did
download the new -dev249 and did cartesian SA, R goes down to 37 but
R-free (50) did not. 

Any suggestions?

    From:
Ralf W. Grosse-Kunstleve <rwgk at cci.lbl.gov>

    To:
    phenixbb at phenix-online.org

    Sent: Tue, December
8,
2009 3:34:09 PM

    Subject: Re:
[phenixbb] refinement

> I do have very low resolution data (around 4 ang), what are

> the efficient way of doing refinement, either rigid body alone or

> rigidbody and tls or individual with group_adp. I did both, but not

> much significant changes in Rfree stays around 48/50%.

You could also try Cartesian or torsion-angle simulated annealing.

I'd try both. In my experience Cartesian SA often works better even

at low resolution.

If you get errors running torsion-angle annealing, please try

the latest nightly build (dev-249) since I've fixed several

problems since the 1.5-2 release.

> Also I do have other questions

>    

> 1. While doing rigid body refinement, phenix complaint about the

> special position and could not perform rigid body refinement,

> whereas individual site refinement is working fine. I do have to

> delete the atom in special position for rigid body refinement?

You could use

  sites.rigid_body = ...

to select the bodies you want to refine. The rest (including your

atom on the special position) will not move.

Ralf

_______________________________________________

phenixbb mailing list

    phenixbb at phenix-online.org

    http://phenix-online.org/mailman/listinfo/phenixbb

_______________________________________________
phenixbb mailing list
phenixbb at phenix-online.org
http://phenix-online.org/mailman/listinfo/phenixbb

  Keep your friends updated— even when you’re not signed in.

_______________________________________________
phenixbb mailing list
phenixbb at phenix-online.org
http://phenix-online.org/mailman/listinfo/phenixbb

_________________________________________________________________
Windows Live: Friends get your Flickr, Yelp, and Digg updates when they e-mail you.
http://www.microsoft.com/middleeast/windows/windowslive/see-it-in-action/social-network-basics.aspx?ocid=PID23461::T:WLMTAGL:ON:WL:en-xm:SI_SB_3:092010
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20091211/da76f2eb/attachment-0003.htm>