Hi Yang Li,
- the data-to-parameters ratio is not good for individual anisotropic
ADP refinement at this resolution;
- I'm wondering why TLS refinement increases the Rfree... How did you
select the TLS groups? Did you try to use TLSMD for this (thanks Ethan,
it produces TLS groups selections ready to use in phenix.refine);
- is there any NCS? if so, use it.
You can also try to use this server to correct for data anisotropy:
http://www.doe-mbi.ucla.edu/~sawaya/anisoscale/
Cheers,
Pavel.
On 5/20/2008 10:00 PM, yang li wrote:
Hi,
I have a structure with 3 different resolutions, 2.3A, 2.4A, 2.5A,
the qualities seem normal, not good but also not too bad.
The B factors along a,b,c axis have notable difference, for example
B(a)=80, B(b)=30, B(c)=20. We used molecular
replacement to solve the structure. For the 2.3A data, the final Rfree
is 0.265 from phenix.refine without tls since tls will
increase the Rfree much. But for the 2.4A data, the Rfree wonnot low
down to 0.32, though the map seems not bad(with
only a few solvent atoms). And for the 2.5A data the Rfree is even
higher than 0.4. For all of them I used thinshell and
followed the same procedure:
MR-->rigid body refinement-->restrain
refinement-->phenix.autobuild-->manually check-->phenix.refine(ordered
solvent)
And autobuild can always build more than 80% residues with mostly side
chains.
This is not a big structure with no more than 1000 residues in 2
molecules. I wonder why the R values keep so high.
Do I need to run anisotropic refinement for such resolutions? Or any
other reasons?
Thanks!
