[phenixbb] Jumping Rfree upon bulk solvent modeling

[Pavel Afonine]

Hi Tobias,

> Dear PHENIXBB community,
>
> i have come across a strange behavior in Phenix:
>
>   * I copied Rfree-flags from a dataset with very slightly different
>     unit cell
>
>   * Refined the previously refined model against the new data
>
>   * Rfactors in first round were good
>
>   * Second round in refinement Rfree was up by 10 % just from the
>     start during bulk solvent modeling and scaling
>
>   * In another round Rfree was high in the beginning, and then dropped
>     by those 10 % during bulk solvent modelling and scaling.
>
>   * In yet another round Rfree started 10% lower than the previous
>     round during bulk solvent modeling in the first cycle and then it
>     jumps up again in the second cycle.
>
> What could be wrong?
>

If I take a random structure from PDB:

phenix.fetch_pdb 1f8t --mtz

and run a default phenix.refine job:

phenix.refine 1f8t.{pdb,mtz}

then look at the log file, I see:

For the first macro-cycle:

(...)
                    start: r(all,work,free)=0.3064 0.3018 0.3435 
n_refl.: 27466
        re-set all scales: r(all,work,free)=0.3064 0.3018 0.3435 
n_refl.: 27466
          remove outliers: r(all,work,free)=0.3061 0.3016 0.3435 
n_refl.: 27464
  overall B=0.38 to atoms: r(all,work,free)=0.3096 0.3050 0.3470 
n_refl.: 27464
bulk-solvent and scaling: r(all,work,free)=0.1839 0.1798 0.2213 n_refl.: 
27464
          remove outliers: r(all,work,free)=0.1839 0.1798 0.2213 
n_refl.: 27464
(..)

Here:
- "start" means no any solvent modeling and scaling was yet done;
- last line shows R values after all corrections.


Now, for the second macro-cycle we see:

()
                    start: r(all,work,free)=0.1798 0.1754 0.2195 
n_refl.: 27464
        re-set all scales: r(all,work,free)=0.3200 0.3151 0.3607 
n_refl.: 27464
          remove outliers: r(all,work,free)=0.3200 0.3151 0.3607 
n_refl.: 27464
overall B=-2.56 to atoms: r(all,work,free)=0.2961 0.2912 0.3366 n_refl.: 
27464
bulk-solvent and scaling: r(all,work,free)=0.1783 0.1739 0.2183 n_refl.: 
27464
          remove outliers: r(all,work,free)=0.1783 0.1739 0.2183 
n_refl.: 27464
()

Here, "start" shows values obtained by the end of 1st macro-cycle, 
second line (re-set all scale) shows values after all scales were 
discarded, and the last line shows values R values after all scales were 
re-calculated again.

So.. if this is what you observe, then there is nothing wrong. 
Otherwise, please send me the log file and I will have a closer look.

Pavel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20190108/b7577ffd/attachment.htm>