Hi Everyone,
Thanks so much for all of your responses! It is encouraging that when I look at the b factors of the protein atoms around my waters/ligands, the b factors track well. All of my lower b factor waters seem to be hydrogen bonding with low b factor residues. I modeled in a few more sodiums and a chloride, which seem reasonable. I tried modeling in some potassiums, but the bond distances seemed to be more consistent with sodium. Now the lowest b factor seen for any water or ligand is around 30 and the average is 40. This seems reasonable especially if you look at my distribution of b factors (the atomic properties tab in Phenix Refine). My overall b factors are spread between 13 to 130 with an average of 60, but I seem to have a bimodal distribution. My first peak is around 40 and my second peak is around 85. I think I’m just modeling the waters present in the more ordered part of my structure.
To Kay, thank you for bringing that paper to my attention. I don’t think it is reasonable to expect that I could possibly get better R factors given the resolution of my data. I have some data at 2.6A, but because of my anisotropy and ellipsoidal truncation, the majority of my data is at much lower resolution (3.5A and 3.7A in the other two dimensions). This particular combination of space group and twin law was the only one which reduced my R factors (I tried about six different combinations), which is encouraging. But yes, I agree that my poor data quality is contributing to the odd spread of B factors.
Thanks again to everyone who took the time to think about my problems.
Best,
Jessica
On Jun 2, 2016, at 7:43 AM, Kay Diederichs <[log in to unmask]> wrote:
>
> Hi Jessica,
>
> those "relatively decent R-values of 28%/31%" may not really indicate that your model is good. Take a look at http://strucbio.biologie.uni-konstanz.de/ccp4wiki/index.php/R-factors#what_kinds_of_problems_exist_with_these_indicators.3F . It cites Phil Evans and Garib Murshudov's paper http://journals.iucr.org/d/issues/2013/07/00/ba5190/index.html which shows that refining a perfect model against pure noise for a perfect twin will give you R-values of 29.1%. So I'd expect a good model to give much lower R-values against data that are better than noise.
>
> In light of this, it is maybe not a surprise that the solvent B factors appear weird.
>
> best wishes,
>
> Kay
>
> On Wed, 1 Jun 2016 20:48:41 +0000, Jessica Bruhn <[log in to unmask]> wrote:
>
>> I was wondering if anyone had any explanations or tips for fixing things. This dataset was highly anisotropic, so I ellipsoidally truncated
>> my data (2.6A, 3.5A and 3.7A in each direction). This dataset is also twinned with a 49% twin fraction. Despite all of that, I was able to build
>> a relatively decent structure (28%/31%) that agrees well with the same protein in a different space group.
>
>
|