[phenixbb] Challenging Refinement at 5 A Resolution

Erik Walinda walinda.erik.67w at st.kyoto-u.ac.jp
Thu Feb 27 23:59:49 PST 2014

Dear community,

This is my first adventure in crystallography and it seems a rather
challenging one.
I obtained at dataset at 5.06 A resolution of a 600 residue protein.

3 domains of the protein are already solved by earlier x-ray
crystallographic studies.

While trying to grow better diffracting crystals and waiting for the next
chance to shoot crystals at the synchrotron, I am trying to fit the already
solved domains into the electron density of the low data set.

Although I tried phaser using the solved structures, I never got a TFZ
higher than ~ 6.4.
Interestingly, Balbes constructed a model which does not contain any of the
already solved domains, but, when using the Balbes output in Phaser, I
obtain a TFZ score of 11 and an LLG of 330. (4 molecules / ASU; P2221)

If this solution is correct, then in principle it should be possible to fit
the already known domains into the structure (I am dreaming of something
like a electron-microscopy-type fitting or the early ribosome structures).
Phenix refinement however raises the Rfree, instead of lowering it. From my
limited knowledge in the field I figured this may be that a) this kind of
refinement doesnt work at this resolution or 2) the MR-solution was not
really a solution.

So I was very excited, when I found the recent papers about DEN refinement
and Phenix-Rosetta.

I am trying DEN refinement first, using the already solved structures as a
reference. The den_refine program nicely creates 2760 DEN restraints.
However, after 1 cycle (about 6 h of CPU time at 30 processors), the Rfree
goes up (Initial: 0.49; final: 0.57). Indeed, the grid search seems not to
be successful, since in the grid search for gamma/w, all Rfree values are
rather similar and the lowest Rfree is 0.54.

So I would like to ask the community, if ...

1) I shouldn't trust the phaser TFZ score too much and maybe it is just no
solution after all?

2) the DEN parameters of phenix.den_refine can be optimized to allow to
refine a super-low resolution structure? (I am using the standard
parameters from the example:

3) 1 cycle is simply not enough and I should let the computation run for a
few days to see, if the Rfree drops down later in the run?

4) Even though Phaser gave me the solution in P2221, may it be that I am
trying to refine it in the wrong spacegroup and that's why I fail?

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20140228/632badc2/attachment.htm>

More information about the phenixbb mailing list