As follow up, the best solution to this problem that I found at the moment is as follows (thanks to Oleg for suggestions/fixes).

1. Set occupancy of the first orientation to 0.5.
2. Set altloc of the first orientation to 1.
3. Set occupancy of the second orientation to 0.5.
4. Set altloc of the second orientation to 2.
5. Rename all the chain IDs of the second orientation (I used lowercase - i.e. A->a, B->b, etc).
6. If you need to keep two orientations close (I do in this case), define NCS groups as follows
refinement.pdb_interpretation.ncs_group {
  reference = chain A
  selection = chain a
}
7. Also, define the occupancy groups as
refinement {
  refine {
    occupancies {
      constrained_group {
        selection = chain A
        selection = chain a
      }
    }
  }
}

This works with phenix.refine.  I've seen some adverse effects with coot - basically, LSQ/SSM no longer works.  Obviously, splitting the full model into separate orientations helps.  And finally, lowercase chains probably need to be renamed back to uppercase for PDB deposition.

Cheers,

Ed.

On Thu, Nov 1, 2018 at 2:15 PM Edwin Pozharski <pozharskibb@gmail.com> wrote:
Is there some way to make phenix.refine to ignore clashes for specific pair(s) of chains?  I am dealing with a rare case of twinning where same protein molecule exists in two orientations in the same unit cell.  I tried just to duplicate the molecule and set initial occupancy to 0.5, but refinement fails (as I expected) saying that there are too many close contacts.  Ideally, I would like to proceed with ignoring close contacts between, say, chains A/B.

I am aware that one can exclude groups of residues from symmetry clashes, but these two copies are in the same cell.  I guess I could try shifting one into the next unit cell and then excluding symmetry clashes, but even assuming that will work it is a rather ugly hack.

The alternative is probably to define the second orientation as an alternative conformation.  Aside from aesthetic objections, I don't want to lose the ability to restrain NCS - I need two orientations to be identical structures, and I am not sure yet whether I can use NCS groups defined via alternative conformers.

Cheers,

Ed.