[phenixbb] differences in refinement stats between phenix-dev-2219 and -1.9-1692

[Pavel Afonine]

Hi Ezequiel,

version: dev-2247: Final: r_work = 0.1753 r_free = 0.2040
vs
version: dev-2247: Final: r_work = 0.1663 r_free = 0.2017

look the same in a sense that if you 100 refinements each one starting 
with a slightly perturbed model you will get 100 slightly different 
refined models with both, Rwork and Rfree, varying withing about 1% at 
this resolution. Having said that I can postulate that r_work = 0.1753 
is the same as r_work = 0.1663, and r_free = 0.2040 is the same as 
r_free = 0.2017.

For some discussion see pages 75-77 here:
http://phenix-online.org/presentations/latest/pavel_refinement_general.pdf
and
"Improved target weight optimization in phenix.refine" here:
http://phenix-online.org/newsletter/CCN_2011_07.pdf

Now, before I proceed commenting, question: did you follow exactly the 
steps 1)-2)-3) that I outlined in my previous email? Otherwise it does 
not make much sense to compare results obtained using different 
refinement settings.

Pavel


On 12/15/15 12:00, Ezequiel Noguera wrote:
> Thank you very much, Pavel. I don't save H's when I modify the model 
> in coot (the input PDB file that I sent you was the output from coot). 
> I did the refinement again, now in version 2247, without NCS. I got 
> the same results, higher R factors and RMS deviations for the newer 
> version. I got better results when I used "update waters" (both in 
> Rfactors and RMS). Below, I list the statistics for refinement without 
> updating waters. Significant differences remain. Could it be any 
> difference in settings of default parameters between versions?
> version: dev-2247
> Start: r_work = 0.1687 r_free = 0.2011 bonds = 0.046 angles = 1.321
> Final: r_work = 0.1753 r_free = 0.2040 bonds = 0.030 angles = 0.593
> version: 1.9-1692
> Start: r_work = 0.1651 r_free = 0.2018 bonds = 0.013 angles = 1.409
> Final: r_work = 0.1663 r_free = 0.2017 bonds = 0.011 angles = 1.321

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20151215/e3b08708/attachment.htm>