[phenixbb] DNA double helix with many molprobity bad clashes

[Nathaniel Echols]

On Wed, Nov 21, 2012 at 5:12 AM, Benda, Christian <benda at biochem.mpg.de> wrote:
> - why is phenix.refine not taking care of these clashes (i.e. optimizes the structure such that they go away)?. Am I giving experimental contribution to high a weight  (I have  "optimize x-ray/stereochemistry" checked)?

Each restraint term needs to be balanced against the rest of the
restraints and the X-ray target, and there is no guarantee that all
parameters will refine to reasonable values.  In fact it is remarkably
easy to come up with strained conformations with bad clashes which
can't be fixed without deleting parts of the model and rebuilding from
scratch.  (This is partly because of the way our minimizer works -
much more aggressive sampling like Rosetta uses often does a better
job getting rid of clashes, at the expense of significantly longer
runtimes.)  In your case I'm not sure what would be the explanation -
two other possibilities include:

- something pathological about where the hydrogens are placed, e.g.
inverted chirality which causes them to clash (this can usually be
fixed by removing and re-adding them)
- something specific to nucleic acids, where the geometry is
traditionally not as well restrained in refinement

Do you mean that the ribose hydrogens are clashing with each other, or
are there other atoms involved in the contacts?  I'd be interested in
seeing the model and maps if you're willing to share them (off-list,
of course!).

> - if so, can I (and how) restrain geometry for the DNA a little more but leave the protein alone (and does that make sense)?

Currently there is an option to apply a constant scale to covalent
restraint terms for an arbitrary atom selection, but it doesn't work
for non bonded interactions yet (this may be problematic anyway).  It
is possible to set the weight on the nonbondeds much higher overall
though - the nonbonded_weight parameter controls this.  You could set
this maybe as high as 200 before the refinement really starts to
choke, but we haven't tested this very much other than to make sure
that the default is somewhat sensible.

I do have an unrelated suggestion however: the nightly builds of
Phenix have completely new hydrogen parameters for Reduce, Probe, and
the geometry restraints, and their handling in refinement has changed
significantly.  So I would recommend trying this:

1) download and install the latest nightly build (currently dev-1218)
2) remove and replace existing hydrogens in the model
3) re-run the refinement with weight optimization (definitely a good
option to use at this stage)

If you still see the same clashes there is something else wrong.

-Nat