[phenixbb] Unrealistic number of waters found by phenix?

Pavel Afonine pafonine at lbl.gov
Thu Feb 17 11:08:32 PST 2011

  Hi Ina,

I'm not sure why you think it is unrealistic.. The solvent content in 
macromolecular crystals may reach 80% or so. Yes, most of it is 
disordered, but the line is pretty subtle - I'm not sure you can 
reliably say how many waters you will find. Not to forget - the amount 
of water you will fins also depends on data and current model quality.
Just to mention: phenix.refine has fairly thorough protocol for updating 
the ordered solvent model during refinement process. Updating the 
ordered solvent model involves the following steps:

1) Elimination of waters present in the initial model based on 
user-defined cutoff criteria on ADP, occupancy and inter-atomic 
distances (water-water, macromolecule-water), (2mFobs-DFmodel) density 
values at water oxygen centers, map correlation coefficient values 
computed for each water oxygen atom.

2) Location of new peaks in (mFobs-DFmodel) map, with following 
filtering of these peaks by their height and distance to other atoms. 
The remaining peaks after filtering are considered as new water oxygens. 
It is possible to specify which type of ADP for newly added waters will 
be refined: isotropic or anisotropic.

3) Depending on the refinement strategy (typically at high resolution), 
individual isotropic or anisotropic B-factors of newly added water 
molecules are refined prior the refinement of all other parameters. This 
is important to do since the newly placed waters have approximate values 
of B-factor (which is usually the average B) and if a large number of 
new waters is added at once this may significantly increase the 
R-factors at this step and slow down the refinement convergence. The 
higher resolution, the stronger the effect.

4) Unlike macromolecular atoms that are ‘wired’ to each other through 
the geometry restraints, the electron density is the only force to keep 
the already added water in place, and occasionally it may happen that 
the density peak is not strong enough to keep water in it and the water 
may drift away from the peak during refinement. Specific algorithm 
implemented in phenix.refine that prevents this from happening.


On 2/16/11 11:52 PM, Ina Lindemann wrote:
> Hi,
> I am refining a crystal structure of a protein with 198 amino acids at 
> 1.59 A resolution.
> I have used the function Update waters in phenix. After visual 
> inspection of the found water molecules I deleted some of them, but 
> there are still 334 water molecules left which show reasonable 
> electron density. Many of them interacts just with other water 
> molecules and not with the protein.
> I calculated the average B of water molecules = 26.3 and protein = 12.6.
> Now I am wondering if the number of water molecules and their average 
> B value is to high with respect to the protein and would be very 
> pleased about your answers.
> With best regards,
> Ina

More information about the phenixbb mailing list