[buster-discuss] Restrained geometry

Clemens Vonrhein vonrhein at globalphasing.com
Wed Feb 7 15:56:35 CET 2018


Dear Xavier,

On Mon, Feb 05, 2018 at 01:04:36PM +0100, xgrcri at ibmb.csic.es wrote:
> Dear Buster List,
> 
> we are currently working with a very large low-resolution structure
> and using BUSTER/TNT for refinement and map calculation.

What constitutes "low-resolution" in your case? Different people/labs
have a different idea about this ... 3A might be "high" for some and
very "low" for others ;-)

What can also be useful to know:

 * Do you have non-crystallographic symmetry (NCS) within your
   structure, i.e. multiple identical molecules within the asymmetric
   unit?

 * Was your structure determined by experimental phasing, molecular
   replacement or simply re-using an existing (isomorphous) known
   model?

   How homologous to your own structure/sequence are your starting
   models (if used)? And at what resolution were those refined (in
   comparison to your own low-resolution structure)?

> Given the limitation of the resolution, we would be very interested
> in applying enhanced rotamer, geometric (for bond lengths and
> angles) and Ramachandran restraints,

Good idea - as long as you are aware that e.g. imposing Ramachandran
restraints will invalidate the Ramachandran plot/statistics as a
validation tool. BUSTER currently doesn't provide this functionality.

> as the evaluation with MOLPROBITY after each refinement cycle gives
> a very high number of outliers

Are these large numbers in absolute terms or also as a percentage of
the total? If you have 50 outliers for 500 residues it would be
worrying - but if you have 5000 residues it seems less severe.

> despite a good overall Clashscore (6.14; 100%) and Molprobity Score
> (2.63; 96%).

A high percentile among structures of a similar (hence low) resolution
is not a convincing quality metric of course. Different Molprobity
versions also return slightly different clashscores.

Obviously, the goal is not necessarily to refine a model against the
Molprobity statistics, but rather against the X-Ray data ;-)

Have you looked at the Ramachandran and/or rotamer outliers? There are
in principle two situations:

 * These residues have clear density that forces them to be an
   "outlier". In that case you could try and check if e.g. a slightly
   different rotamer also fits: at lowish resolution, a lot of
   side-chains can easily be flipped around (ILE, LEU, VAL etc).

 * These residues have no density and they get pushed around by other
   forces (e.g. bad contacts) without any X-Ray term holding them in
   place. You could consider truncating them down to Cbeta or removing
   whole stretches completely.

Other points to consider:

 * Have you tried (re)processing your data e.g. with autoPROC [1]
   and/or using the STARANISO server [2] to check for anisotropy? If
   your data has severe anisotropy that seems to sometimes help in
   refinement and providing clearer density after BUSTER.

 * If you started from a well-refined, high(er) resolution model it
   can easily happen that refinement "looses" the good geometry
   through refinement, because the X-Ray data is not providing the
   additional information. In that case you could try so-called
   LSSR targeting [3] to stay close to your good starting model.

 * If you have NCS you should always use LSSR restraints via the
   -autoncs flag.

   You could also use "-autoncs_noprune" so that all residues will be
   within the NCS relation (no automatic pruning of significant
   outliers) - at least at the beginning.

   There is another flag (-sim_swap_equiv) that tries to re-introduce
   NCS relations for symmetrical side-chains (ASP, GLU, PHE, TYR and
   ARG) where a 180-degree rotation won't change the chemistry, but
   only atom names. Such a change in atom names has a knock-on effect
   when comparing NCS-related copies in your model.

   This also means that you should always "Cancel" the suggestion by
   Coot to "fix" nomenclature upon loading of your PDB file: one
   cannot be sure that this fix will alter atom naming of all
   NCS-related copies of the same residue in a consistent way and can
   therefore lead to a break-down of NCS-relation ... which
   "-sim_swap_equiv" can again try to fix, but it is better to not
   introduce errors than to try and fix them later.

   If you do manual rebuilding: make sure to do this in an NCS-aware
   manner, eg. a rotamer change in one chain needs to be propagated to
   all other chains (unless there is clear indication of differences
   supported by density).

> Is there a way to impose such restraints?

Anything you can describe in terms of distances [4], angles [5] or
torsions [6] can be added to the restraints. We've done that for
RNA/DNA base-pairing or secondary-structure restraints, although this
is not (yet) in our distribution.

Finally, the latest release of BUSTER contains tools from the PDB_REDO
[7] suite to help with model rebuilding. This could be potentially
helpful in your case too [8].

Cheers

Clemens, Claus & Gerard (for buster-develop)

[1] https://www.globalphasing.com/autoproc/
[2] http://staraniso.globalphasing.org/
[3] https://www.globalphasing.com/buster/manual/gelly/manual/gelly4A.html#target
[4] https://www.globalphasing.com/buster/manual/gelly/manual/gelly5.html#dist
[5] https://www.globalphasing.com/buster/manual/gelly/manual/gelly5.html#utilangle
[6] https://www.globalphasing.com/buster/manual/gelly/manual/gelly5.html#utiltor
[7] https://pdb-redo.eu/
[8] http://www.globalphasing.com/buster/ReleaseNotes/ReleaseNotes-BUSTER_snapshot_20171219.txt


More information about the buster-discuss mailing list