On Thu, Nov 7, 2013 at 3:50 PM, Ryan Spencer <rspencer@uci.edu> wrote:

I have found that phenix.model_vs_data does not deal well with a perfect twin data set. The refined data is:

r_work=0.2013 r_free=0.2298 (in PDB file)

r_work=0.2815 r_free=0.3006 (from phenix.model_vs_data)

There is one merohedral twin operator: -h-k,k,-l (48%) and 6 very small (<1.0%) pseudo-merohedral twin operators.

Yes, the automatic twinning detection is one of the weak points of the current Table 1 program (and the enhanced MolProbity validation in Phenix); a more flexible approach will be available soon. However, if you have a case where you're certain the structure is twinned (and these R-factors do support that hypothesis) but Phenix is missing it, we'd be interested in seeing the model and data. It may just be a limitation of the twinning, but occasionally our code trips on corner cases, or it could be something stranger misdiagnosed as twinning.

-Nat