I am refining the crystal structure of a protein that is around 20 kDa. I am getting towards the point where most of the validation statistics and residual look nice: resolution 1.85Å, Rfree=19.5% Rwork=15.5%, Angle and bonds RMS below 0.01Å respectively 1.1˚, mol probity score < 1.1, No Ramachandran outliers and over 98% of Ramachandran favoured, no C-beta outliers, very low clashscore. There are a few rotamer that still need a little bit of attention. After correcting them manually with Coot I usually follow with a short round of refinement (3 macro cycles) with Phenix.refine using the following parameters:

Real space position (XYZ), TLS, individual B-factors and both X-ray /stereochemistry and X-ray ADP weight optimisation. Unfortunately The R-stats always get worse (e.g. +1% for R-free) while the geometry (angle and bond) marginally improves if at all. The rotamers that I have just corrected are often back to the wrong position. Basically my model gets worse. I have played around with the different parameters but I always end up with the same problem. What is the best refinement strategy in this kind of situation? I was wondering if there is a way to tell phenix to make minimal changes to most of the structure and only focus on the area that I have recently modified. I would be keen to learn how to play with Phenix a little more efficiently.

Your help is much appreciated.

Best regards

Guillaume