[phenixbb] problem refining RNA structure at 4Å resolution

Nathaniel Echols nechols at lbl.gov
Wed May 29 08:04:37 PDT 2013

On Wed, May 29, 2013 at 5:53 AM, Benoît Masquida
<b.masquida at ibmc-cnrs.unistra.fr> wrote:
> I solved using MR a low resolution RNA structure taking data up to 4.3Å and a model of a structure with slight differences. However, when I try to refine, Rwork decreases quickly (start 47, final 32 but Rfree remains high (start 48, final 44). The data are pretty good up to 4.0 with I/sigma above 1. The map doesn't point clearly to wrong regions of the model. What can I try to improve the maps and build the correct model?

Standard advice at this resolution:

1. Use reference model restraints.
2. Use NCS restraints if you have NCS (at this resolution it might be
worth trying both the torsion and cartesian parameterizations).
3. Optimize the X-ray/restraint weights.
4. Make lots of omit maps to ensure that you're not being deceived -
model bias at this resolution really is scary.  Also be careful if the
outer resolution shells are incomplete, because Phenix (like Refmac)
will by default fill in missing reflections with F-calc to avoid map
artifacts, but this can make the model bias worse.
5. Use map sharpening - I would recommend doing this interactively in
Coot rather than checking the box in Phenix.

You might also want to experiment with more conservative B-factor
parameterizations (grouped or TLS), although I have yet to figure out
how to do this reliably (and get a better result than tightly
restraint individual isotropic refinement).


More information about the phenixbb mailing list