Hi Damian,
In the other 5 copies, a entire domain (~25kDa) is largely disordered (some have patchy residual density, but not readily interpretable). Altogether, this means that I am missing ~20% of the total protein in the AU. It seems that how you model this much "missing" material could have a significant effect on the final refined model. This is something we have observed a number of times, so I am wondering if anyone can suggest ways to deal with this aside from just leaving the domains out entirely.
Regarding missing or patchy domains, Pavel recently added a feature that should at least improve the phases and refinement behavior
this is still ongoing project that Tom and myself are working on. It still need some work and a lot of testing. You can try it now by using "use_statistical_model_for_missing_atoms=true" in phenix.refine (you need the latest version for this), and there is no documentation yet, sorry. Actually, I was looking for a real case where there is a significant amount of model missing. If you send me the data, your best current model and the sequence, I will use it for testing and optimizing/training the algorithms that we are developing right now. For some details see pages #17-19 here: http://cci.lbl.gov/~afonine/afonine.pdf Pavel.