Hi Nat,

It is enabled by default because it is as equal piece of information as restraints on lengths of covalent bonds, values of covalent angles and all other restraints. It is simply extra information that would be shame to not use actively especially if it comes for free, and is trivial to define unambiguously (unlike Ramachandran restraints, for instance).

I've made it default perhaps a year ago and I've not seen any problem related to this since then. I can demonstrate improvements in low-resolution real-space refinement when using CB restraints vs not using them, for example.
If you can bring an example that proofs your concerns (that is clearly shows that using these restraints is counterproductive) I will be happy to investigate!

As to diagnostics, I think proper use of existing metrics such as local (per atom) 2mFo-DFc and mFo-DFc map fit, along with local standard geometry quality metrics (clashes, deviations from ideal targets, etc) should be totally sufficient to identify problems.

Pavel

On 11/7/14 8:35 AM, Nathaniel Echols wrote:

On Fri, Nov 7, 2014 at 2:08 AM, Pavel Afonine <[email protected]> wrote:

Phenix.refine uses CB restraints by default, so unless you disabled it or your model is very poor, you should not have these outliers after refinement (again, normally).

Why is this enabled by default? CB outliers are diagnostic for other errors in the model, like the scenario of the partially-split sidechain I described. This is the reason why they're the only chiral centers flagged in MolProbity. If the rest of the atoms in the residue are compatible with a proper CB position, the standard geometry restraints should be sufficient to keep it there. If not, adding more restraints will either introduce other errors into model geometry, or mask the actual problem with the model. Either way, this is not actually going to lead to better structures.