[phenixbb] optimizing weights

Nathaniel Echols nechols at lbl.gov
Wed Apr 20 12:34:21 PDT 2011


On Wed, Apr 20, 2011 at 12:21 PM, Alexandra Deaconescu
<deacona at brandeis.edu> wrote:
> I would like to optimize my weights during Phenix refinement. I am running
> Phenix 1.6.4-486 using the gui, and I have checked the "optimize weights",
> but I cannot see the results of the optimization in the output, so I do not
> know the relationship between Rfree/geometry and wc and wxc_scale (I might
> not want to pick the automatically selected wxc_scale for the lowest Rfree,
> but would like to find a reasonable values to give me good geometry). Some
> earlier posts indicated that this is not output in the log file, but I am
> running a more recent version...

It definitely should be in the log file, although it probably doesn't
explicitly state "optimizing weights" anywhere.  (The GUI doesn't
display any information about weight optimization other than the log
output - but I could be convinced to add something.)  It won't run
every macro-cycle by default (this is a separate option); if you run 3
macrocycles, I think the weights get optimized in the 2nd.

> Same question for the ADP refinement.... I am working at 2.8A and my B
> factors are high (higher I think than what one would expect at this
> resolution).

How much higher?  What does Xtriage say the overall isotropic and
anisotropic B-factors for the data are?  (These aren't perfectly
correlated with the atomic B-factors, but they should track them to
some degree.)  The POLYGON graph in the validation output is helpful
here - there is a very wide distribution of B-factors in the PDB.  The
attached image shows the mean backbone B-factor at 2.7-2.9A - I'm not
sure I believe some of the extreme values, but it's clear that a mean
B above 100 is still possible.

PS. You should upgrade to version 1.7.

-Nat
-------------- next part --------------
A non-text attachment was scrubbed...
Name: b_distribution.png
Type: image/png
Size: 36520 bytes
Desc: not available
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20110420/c57744ff/attachment-0001.png>


More information about the phenixbb mailing list