At what resolution is (individual,group,one per residue, two per residue) appropriate? was Re: [ccp4bb] Structure Determination combining X-ray Data and NMR

From: Francis E Reyes
Date: 6 January 2012 18:18


Pete brings out two concerns: [1] B-factor refinement being stable, [2] over-fitting at low resolutions.  Here I'll suggest low resolution to be in the 3-4A range.

I've seen the following question asked:  At what resolution is (individual,group,one per residue, two per residue,overall) appropriate?

One common response is "try a number of different B-factor refinement protocols, use Rfree as a guide to determine which one is appropriate".  I guess this is along the lines of "B-factor refinement being stable". Seems reasonable to me.

However is this approach sufficient justification to address "over-fitting at low resolution"?



I'd welcome any thoughts on this ...
F

On Jan 6, 2012, at 11:00 AM, Pete Meyer wrote:
>>
>
> B-factor refinement being stable is one thing; quieting my paranoia regarding over-fitting at low resolutions is another.
>
> Thanks for pointing this out to me - I'll have to check out the details of how phenix handles it, and give it a try.
>
> Pete



---------------------------------------------
Francis E. Reyes M.Sc.
215 UCB
University of Colorado at Boulder

----------
From: Ed Pozharski

My personal opinion is that the individual B-factor refinement with
restraints proper to the resolution is the only thing you will ever
need.  This thread may be interesting to you.

http://www.mail-archive.com/ccp4bb@jiscmail.ac.uk/msg14133.html
As a general rule, the R-free should never be the only criterion.
Additional criteria will vary depending on what alternative protocols
you are investigating.  To try to generalize, I'd say that improved
unbiased fit (via R-free) should be combined with the requirement that
model parameters are physically meaningful.  For example, the
two-per-residue B-factor model fails miserably on that account, and my
sincere wish is that it's use in modern refinement software will be
obfuscated to the extent needed to render it unusable.

Cheers,

Ed.

--
"I'd jump in myself, if I weren't so good at whistling."
                              Julian, King of Lemurs

----------
From: Pavel Afonine


Francis,

I recognize my advice along those lines above -:)

Although now I would say (assuming phenix.refine): "Try individual ADP refinement first, and if it fails then try group ADP refinement". By "fails" I mean overfitting as judged by Rfree and Rfree-Rwork.

I don't know what that means :  "B-factor refinement being stable".

I think yes.

Pavel