Hi Scott,
I am trying to perform refinement on a model with three chains (two protein and one DNA... plus water, Cl, and a few other solventy things). The B factors for chain A are generally better than for the other two and I am thinking that the resolution (1.8Å) warrants attempting anisotropic individual adp refinement for at least chain A, and then using TLS for the other two.
You can do it in phenix.refine.
So my strategy is:
*individual_sites *individual_adp *tls *occupancies
adp.individual.isotropic = "not Chain A" adp.individual.anisotropic = "Chain A" adp.tls = "Chain B" adp.tls = "Chain D"
This looks good too.
Previously the model had only been refined with individual isotropic adp (no TLS). The starting model has an r_work=0.2079 and r_free=0.2427. With this new aniso/TLS strategy in the first macrocycle when it gets to the initial TLS refinement it completely blows up with r_work and r_free both going up to 60%
I have no idea why this happens. The only way I can answer is to reproduce this problem myself. Could you please send me the data, model and the exact command you used (including parameter files, if any). Pavel.