[phenixbb] R-free set for low resolution

[Nathaniel Echols]

On Mar 30, 2010, at 7:05 AM, r n wrote:
> Follow up the shaking the coordinates, I was end up using different R-free set. 
> 
> I do have a data is at very low resolution, so only possible to do rigid body refinement. 
> 
> How can I  avoid over fitting or model bias due to the different R-free set using phenix.

If you can't do restrained refinement, your options are basically limited to what Pavel and I already posted: shake the coordinates and regularize, or run simulated annealing without the X-ray term.  (We probably need to create a separate command for this.)  These have the disadvantage of potentially introducing distortions that are far from the native state of the protein in the crystals, and which won't be corrected by rigid-body refinement.   Resetting all of the B-factors to something appropriate (the Wilson B, for instance) is a good idea, especially since it won't distort the model.

However, if you have access to the original R-free set, you should really use that instead.  My (uninformed) guess is that if you're only using rigid-body refinement, the probability of overfitting badly is low anyway, but you will still have some model bias.

(PS. I would recommend including TLS refinement, with each domain or chain as a separate group.)

-Nat

--------------------
Nathaniel Echols
Lawrence Berkeley Lab
510-486-5136
NEchols at lbl.gov