OK, let's experiment.

I'm taking a porin model and I know that the data is affected by twinning (porin.pdb and porin.mtz).

1) Let's run Xtriage to find out the twin law:

phenix.xtriage porin.mtz

and it gives me this:

Statistics depending on twin laws
-----------------------------------------------------------------------------------------------------------------
| Operator                                                 | type | R obs. | Britton alpha | H alpha | ML alpha |
-----------------------------------------------------------------------------------------------------------------
| -2/3*h-1/3*k+2/3*l,-1/3*h-2/3*k-2/3*l,2/3*h-2/3*k+1/3*l  |  PM  | 0.410  | 0.079         | 0.084   | 0.022    |
| -h,1/3*h-1/3*k+2/3*l,2/3*h+4/3*k+1/3*l                   |  PM  | 0.419  | 0.070         | 0.069   | 0.022    |
| -1/3*h+1/3*k-2/3*l,-k,-4/3*h-2/3*k+1/3*l                 |  PM  | 0.415  | 0.070         | 0.075   | 0.022    |
| -h,2/3*h+1/3*k-2/3*l,-2/3*h-4/3*k-1/3*l                  |  PM  | 0.420  | 0.068         | 0.069   | 0.022    |
| 1/3*h+2/3*k+2/3*l,-k,4/3*h+2/3*k-1/3*l                   |  PM  | 0.415  | 0.074         | 0.077   | 0.022    |
| -1/3*h-2/3*k-2/3*l,-2/3*h-1/3*k+2/3*l,-2/3*h+2/3*k-1/3*l |  PM  | 0.413  | 0.078         | 0.079   | 0.022    |
| h,-h-k,-l                                                |   M  | 0.195  | 0.292         | 0.315   | 0.304    |
-----------------------------------------------------------------------------------------------------------------

so I'm still puzzled which twin law to choose from the above list, so I run phenix.model_vs_data:

phenix.model_vs_data porin.{mtz,pdb} > model_vs_data.log

phenix.model_vs_data will score the possibilities: it internally runs Xtriage to get the twin law and then it computes the R-factors with and without considering twining and based on this it tells you what you have.

Ok, phenix.model_vs_data outputs:

(...)
  Data:
    twinned                 : h,-h-k,-l
(...)

and I take "h,-h-k,-l" as my twin_law.

2) Next, I' going to run two refinements - with and without considering twinning:

phenix.refine porin.{pdb,mtz} output.prefix=twin twin_law="h,-h-k,-l" --overwrite

and

phenix.refine porin.{pdb,mtz} output.prefix=no_twin

Obviously, considering twinning is a good idea:

twin_001.pdb:
Final: r_work = 0.1207 r_free = 0.157

no_twin_001.pdb:
Final: r_work = 0.2038 r_free = 0.2475

3) Now let's play with SA omit maps. The file "porin.pdb" contains a ligand "C8E".

I'm omitting the ligand from twin_001.pdb file (twin_001_NoLigand.pdb) and start SA refinement:

phenix.refine porin.mtz twin_001_NoLigand.pdb  output.prefix=twin_sa_omit twin_law="h,-h-k,-l" simulated_annealing=true --overwrite

and the result is:

twin_sa_omit_001.pdb:
Final: r_work = 0.1224 r_free = 0.1650

which is slightly worse than

twin_001.pdb:
Final: r_work = 0.1207 r_free = 0.157

but is not surprising at all since:

- By omitting the ligand we made the model worse. One should not expect the omit map look better than the regular map;

- SA is good for correcting large errors. This is why it does very good at the beginning of refinement, but it rather harms when you run it with a good close-to-final structure.

The maps twin_sa_omit_001_map_coeffs.mtz and twin_001_map_coeffs.mtz look almost identical, and the mFo-DFc omit map shows the ligand.

>From this I conclude that the program works as expected and does not contain obvious bugs. If you experience a problem with your particular structure, then I think it's case-specific.

All the files mentioned above are located here:
http://cci.lbl.gov/~afonine/twinning_experience/porin/

Let me know if you have any questions.

Pavel.