<div dir="ltr">On Mon, Aug 25, 2014 at 1:12 AM, marinabb phenix <span dir="ltr"><<a href="mailto:marinabb.phenix@gmail.com" target="_blank">marinabb.phenix@gmail.com</a>></span> wrote:<br><div class="gmail_extra"><div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div>I am trying to use DEN refinement as implemented in the Phenix-1.9 release to refine a structure at 3.6 Angstrom resolution. The asu consists of 8 copies of hetero-dimers (of roughly 40 KDa sunbunits each). The structure of one of the subunits has already been solved at about 2 Angstrom, whereas for the other no homologous model is available. After automatic + manual building, I have now a model with Rfree = 28 / R = 23, with some protomers better built than others. I have used this pdb model as input for DEN refinement, adding the known higher-resolution structure as a (partial) reference model, and ncs restraints previously found (and run a DEN parameters optimisation). The final DEN Rfee got significantly worse.</div>
</div></blockquote><div><br></div><div>I wouldn't expect DEN to be much help here; the R-free is already quite respectable for this resolution, which means that most of the model is (probably) in approximately the correct location. DEN is good for scenarios where the model needs to move a lot. I would focus on improving the geometry and completeness instead.</div>
<div><br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div>I would appreciate any advise for a better DEN strategy to be adopted in this case, if possible. In particular I was wondering whether it can be worth building a model for the missing subunit (or part of it) based on structural similarity of the current unrefined 3D structure.</div>
</div></blockquote><div><br></div><div>I've never heard of this being tried before, but it's probably not that much different from what MR-Rosetta is going internally. So you could try running MR-Rosetta in "model improvement" mode (without running the MR). The hybrid Rosetta-Phenix refinement may also be useful, although it doesn't (yet) do the kind of fragment-based rebuilding that MR-Rosetta does.</div>
<div><br></div><div>Two generic bits of advice:</div><div><br></div><div>1) if you're willing to email us the model and data off-list at "<a href="mailto:help@phenix-online.org">help@phenix-online.org</a>", we may be able to think of more specific answers.</div>
<div><br></div><div>2) How did you decide on a resolution cutoff of 3.6Å? You may benefit from pushing it further, for example to the point where CC1/2 falls below 0.5. I don't know what the odds are of this helping, but it rarely hurts, and we've seen enough anecdotal evidence to support making it common practice.</div>
<div><br></div><div>-Nat</div><div><br></div></div></div></div>