[buster-discuss] rejected refleXions in refine version 2.10.2

[Clemens Vonrhein]

Dear Wolfram,

(sorry for the late reply).

On Thu, Apr 06, 2017 at 04:29:15PM -0400, wtempel wrote:
> 01-BUSTER/Cycle-5/LIST.html includes a dead link to a file
> hkl_reject.html

Correct: we remove a fair amount of files that are not really if
interest to users (and even we developers hardly ever miss them unless
we do some hard-core, open-heart surgery on BUSTER).

> Is there another file that would allow me to view an analysis of
> rejected reflexions?

The above file (hkl_reject.html) does not contain rejected reflections
in the sense that you are thinking about: it lists reflections that
are rejected from refinement because their test-set flag marks them as
being in the "free" set.

> In my case, the lower resolution limit changes from 25 Å (hklin) to
> 12 Å (reported in refine.pdb), corresponding to approx. 25
> reflexions.

This is done automatically based on some very (!) poor correlations at
low resolution in the reciprocal space correlation plots. You could
switch off that behaviour using

  AutomaticRestrictLowres="no"

However, it nearly always points to some problems in processing (beam
stop masking etc) or collection (overloads etc).

> The diffraction images may have had some overloaded pixels, and
> scaling statistics show the low resolution shell less complete than
> the overall data set.

That might be the explanation: you now have only (relatively) weak
reflections in the low-resolution range and all strong reflections
required to model the bulk solvent contribution reliably are missing
(resulting in poor low-resolution CCs).

> I would like to confirm that those lowest 25 or so reflexions were
> in fact rejected by buster.

Since BUSTER doesn't reject data based on individual reflections, you
can just look at the data in the range 25-12A, i.e.

  echo "RESO 999.9 12.0" | mtzutils hklin your.mtz hklout lowres.mtz
  mtzdmp lowres.mtz -n -1

> Would the ensuing loss to the quality of my model justify the added
> complexity introduced by possibly merging the high resolution
> (synchrotron) data with an earlier, lower resolution rotating anode
> data set from the same crystal?

Difficult to tell: if you are working on a structure that relies
heavily on good low-resolution data (only low-resolution data
available, disorder, missing domains etc) you will probably see an
effect. The question is if using systematically skewed data at
low-resolution (only weak reflections) is better/worse than not using
the low-resolution data at all.

Trying to combine a high-resolution and low-resolution dataset can be a
tricky business in itself. Doing this for data from different crystals
makes it even more complicated ... and data from different
detectors/instruments will not make it easier.

What is still unclear, missing or disordered in your current model
after BUSTER refinement using data with a low-resolution cutoff of
12A?

And what is the high-resolution limit of your data?

Cheers

Clemens (for buster-develop)