[phenixbb] Adequate size for Free R test set?

[Nathaniel Echols]

On Tue, Aug 3, 2010 at 11:21 AM, Joseph Noel <noel at salk.edu> wrote:

> Thanks so much for the suggestions. They are really helpful! A few
> questions relative to Phenix. Is there a way to check with Phenix the thin
> resolution shells? I created Free Rs in Phenix using the thin resolution
> shells options but removed the 2000 limit and instead used 5%. Maybe I
> over-did it relative to your reply. Just wondering how to check each of
> these thin shells. Alternatively, does Phenix ensure one has an adequate
> number of test reflections in each "thin shell" if this option is set? If
> so, what should I used for the maximum number of reflections and/or %?
>

There isn't a good way to check the shells in the GUI - but all of the
pieces are there, and I've been meaning to add some utilities for examining
reflection files anyway.  For now, on the command line, run this:

iotbx.r_free_flags_accumulation data.mtz

and it will print out something like this:

Number of work/free reflections by resolution:
                                           work  free  %free
  bin  1: 135.8135 -   4.3073 [7861/7861]  7663   198   2.5%
  bin  2:   4.3073 -   3.4187 [7682/7682]  7484   198   2.6%
  bin  3:   3.4187 -   2.9866 [7645/7645]  7444   201   2.6%
  bin  4:   2.9866 -   2.7135 [7605/7605]  7406   199   2.6%
  bin  5:   2.7135 -   2.5190 [7593/7593]  7395   198   2.6%
  bin  6:   2.5190 -   2.3704 [7618/7618]  7416   202   2.7%
  bin  7:   2.3704 -   2.2517 [7535/7535]  7337   198   2.6%
  bin  8:   2.2517 -   2.1537 [7573/7573]  7374   199   2.6%
  bin  9:   2.1537 -   2.0708 [7556/7556]  7354   202   2.7%
  bin 10:   2.0708 -   1.9993 [7547/7547]  7348   199   2.6%
                                  overall 74221  1994   2.6%

This is a real example, using the default settings in the reflection file
editor (20 thin shells for the test set).  It looks much better than what I
had originally done with this dataset five years ago using a different
program, where the last shell had only 0.2% of reflections flagged.
 However, the code that does this isn't particularly sophisticated, so I
would recommend double-checking the output for your data.  I don't think
you're going to do any harm by sticking with 5% and removing the absolute
limit.

-Nat
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20100803/c216aafa/attachment-0002.htm>