[phenixbb] Failed ADP refinement with active TLS and weight optimization

Schubert, Carsten [JRDUS] CSCHUBER at its.jnj.com
Mon Aug 10 11:08:09 PDT 2015


Hi,

I am running into some strange issues with regards to ADP refinement especially when I am putting final touches on a structure for PDB submission. The problem manifests itself most prominent in the B-factor refinement of solvent, but also maps are affected and are becoming noisier, so I assume this problem affects the protein as well.  I checked this against the current release 1.9 and the latest pre1.10 dev version, both versions show the issue. Refinement protocol includes individual_xyz, individual_adp, TLS and weight optimization.
The problem I am running into is that the ADP of waters  only changes by a global offset. The problem is tricky to spot but more evident when looking at solvent placed in coot with the default ADP (30). The ADP changes only by a global offset, no individual refinement takes place. Since the protein is refined by TLS the problem is not as evident, but as mentioned the fofc maps get noisier from cycle to cycle even though no manual change to the geometry was performed. To me an indicator that the B-factors are not refined properly.

The problem can be circumvented by reverting back to isotropic B-factors and then refine with weight refinement for 5-10 cycles. After that TLS can be switched back on for a final polishing cycle. In stubborn cases I had to randomly assign B-factors close to Wilson-B and refine for more cycles to get out of the hole I refined myself into.

Another indicator that something went awry is the weight refinement for the B-factors:

                 ----------Individual ADP refinement----------

    R-FACTORS      <Bi-Bj>  <B>   WEIGHT       TARGETS
work  free  delta                           data restr
20.66 26.39   5.73  5.712  67.871  2.160    5.225
20.66 26.38   5.73  5.712  67.871  0.030    5.225
20.66 26.38   5.73  5.712  67.871  0.125    5.225
20.66 26.38   5.73  5.712  67.871  0.500    5.225
20.66 26.38   5.73  5.712  67.871  1.000    5.225
20.66 26.38   5.73  5.712  67.871  1.500    5.225
20.66 26.38   5.73  5.712  67.871  2.000    5.225
20.66 26.38   5.73  5.712  67.871  2.500    5.225
20.66 26.38   5.73  5.712  67.871  3.000    5.225
20.66 26.38   5.73  5.712  67.871  3.500    5.225
20.66 26.38   5.73  5.712  67.871  4.000    5.225
20.66 26.38   5.73  5.712  67.871  4.500    5.225
20.66 26.38   5.73  5.712  67.871  5.000    5.225

Essentially there is no change in R/Rfree with regards to different weights.

I have not figured out the exact time point when this happens during the course of the refinement but it seems to be coupled to using TLS in combination with ADP refinement and during the final stages of the refinement.

I can provide data offline to reproduce the issue

Cheers,

Carsten
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20150810/2059210a/attachment.htm>


More information about the phenixbb mailing list