[phenixbb] One adp per group gives very high variance between adjacent residues.

Pavel Afonine PAfonine at lbl.gov
Thu Dec 31 10:00:54 PST 2009


Hi Kendall,

>  I am resetting the B factors in both cases. TLS definitely helps 
> (does it every not help?).

The goal of using TLS model is not to reduce R-factor, but is to use 
more physically correct model (this is primary), and eventually reduced 
R-factor (this is secondary) is the reward for doing so; while increased 
R-factor will indicate the incorrect ADP model.

> If poor fit is the problem, why do I not see the variance with 
> individual ADPs? 

Because the ADP restraints are used in case of individual ADP 
refinement, and no restrains are used in case of group B refinement. Did 
you try refining individual ADPs with two scenarios:
- using "wxu_scale=true" to optimize weights;
- manually setting very string ADP restraints ?
I would definitely try both options.

> I still don't think it makes physical sense to have one residue in a 
> helix with an ADP of 100 and the next one with 200, 

I did not say it is correct or makes physical sense. I just tried to 
explain why it happens (the most likely reason).

> which is what I'm seeing with the one per residue setting. With 
> respect thermal motion, the adjacent residues are attached, so it 
> makes sense to me to have their values restrained. 

Copy-paste from my previous email:
"""Imagine one residue that fits density well and its group B-factor 
refines to a good value and the adjacent residue is out of density and, 
no surprises, its refined group B is very high. In this case if we apply 
restrains, we will spoil the group B-factor of the good residue by the 
high B-factor of the bad one, and we will force the high B of the bad 
residue to be unrealistically smaller than supported by the density."""

which is saying differently : If both residues fit density well, then 
applying restraints will not do any harm, and this is not the case 
otherwise.

If in your particular case both residues fit density very good and the 
difference between their refined group B-factors is very large, then the 
problem is somewhere else, and this needs more investigation (which may 
take a day or more). Also, I will need to be able to reproduce it myself 
(I need the data and model for this). I will see what it takes to add 
restrains for group B refinement, at least for experimenting.

Pavel.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20091231/e52dab27/attachment-0002.htm>


More information about the phenixbb mailing list