>The way Phenix.refine refines a model with a high twin fraction is different >(proportionality rules) which can introduce model bias, so your model should >be pretty close to the end product. I recommend refining your structure as >well as you can without inputting the twin law until the very end.
Minor comment: Detwinning is not used in refinement: the target function used is 'twin aware' and there is no need for dewtinning of data until you actually want to look at the maps (water picking etc etc) It is true indeed that when detwinning data using proportionality rules one introduces some bias. The extend of this bias depends on the presence and amount of pseudo rotational symmetry. When no pseudo rotational symmetry is present, I suspect this bias to be high. When a lot of PRS is there, I suspect the bias is less of an issue. I hope that you have assigned your free set with phenix.refine. in that way, you are certain that the symmetry of the free set follows the point group symmetry of the lattice. This should avoid potential pitfalls Eleanor mentioned. HTH P
