Hi,
1. I deleted the whole DNA (947 atoms of DNA against 7543 for protein part) from pdb and ran phenix.refine with simulated_annealing=True. But resulted Fo-Fc in the omitted region is very weak. Is it because of the low resolution and because big part of the model was omitted? This is part of the problem - another possible contributing factor is that the bulk solvent mask will extend into the DNA site, which flattens out the difference density. An alternate way to do this (which should work) is to set the occupancy of the DNA to zero, which will remove any contribution to the scattering, and set the parameter ignore_zero_occupancy_atoms=False, which will prevent the mask from extending into the DNA.
Yes, that's the trick. We should streamline it somehow. However, removing such a big fraction of atoms at once can't be painless: you can't remove arbitrarily large amount of atoms and expect a useful map after that. May be you should split atoms to omit into fractions (like 1/3 or so), and compute OMIT maps for each one fraction, and then combine resulting maps into one. Pavel