On Thursday 08 July 2010 11:03:30 am Parthasarathy Sampathkumar wrote:
> Dear All,
>
> Back ground:
> This is my first experience with a twined dataset. Crystals belong to a
> small domain of 132 aa, out of which ~40 residues appears to be disordered
> (~30 of those from C-terminal and C-term His6 tag).
>
> Initial space group: P3 with unit cell dimensions: 62.507 62.507 55.117
> 90.00 90.00 120.00; Resolution 2.35Angs.
>
> Pointless suggested P321 (space group # 150) as possibility. I determined
> structure by MR with Phaser (1 molecule in ASU). After several model
> building and refinement cycles R and Rfree got stuck at 24.0% and 31.6%
> respectively.

OK.

> Therefore, I considered P3 space group (now Two molecules in the ASU) with a
> twin component.

But now you veer off track. P3 + 0.5 twinning fraction is the same
as P321, except that you have doubled the number of refined parameters
by modelling two copies of the protein rather than one. This is
probably not what you want to do.

> I was only able to add handful residues to the model already
> refined in P321. My current R and Rfree factors are 21.0% and 29.1%,
> respectively for Two molecules refined in P3 space group.

Turning on twin refinement in refmac is always expected to drop R slightly,
regardless of whether the structure is actually twinned or not. I'll leave
it to Garib to explain why this is the case :-) In other words, given that
you doubled the number of parameters and also turned on twin refinement,
a drop in Rfree from .31 to .29 is not impressive.

I don't think that any of the information you present justifies treating the
structure as being in P3.

Instead you should be considering the possibility that the structure really
is in P321, but there is a twin law that makes it appear partially hexagonal.
See the table of possible twin laws for P3(?)21:
   http://www.ccp4.ac.uk/html/twinning.html

> Questions:
>
> 1. H-test in cTruncate suggested a twin fraction of 0.43 for the twin
> operator -h-k, k, -l.

Was this relative to a P3 indexing, or a P321 indexing?

> Where as Refmac5 with Amplitude based twin refinement
> gave an initial value of 0.508 for the same operator.

That's strange. A twinning fraction should never be higher than 0.5
But let's get your true twinning law sorted out first before worrying
about whether refmac should report this as 0.508 or 0.492

Ethan


> Why these values are
> different between cTruncate and Refmac5 (is this because I asked Refmac5 do
> amplitude Twin refinement instead of Intensity based)?
>
> 2. I noticed in Refmac5 log file that twin fractions changes for every cycle
> of refinement. During my most recent Refmac5 run Initial estimate of 0.508
> for -h-k, k, -l operator changed to 0.504 at the end of 20th cycle. The
> corresponding values were 0.519 and 0.529, respectively, in a previous
> round. Since twin estimates were based on measured Intensities (in turn
> amplitudes) why would they change with refinement (am I missing something
> here)?
>
> 3. When I repeated my final round of Refmac5 WithOut Twin Refinement my R
> and R-free factors are 22.9% and 28.6%, respectively, which also appears to
> be OK for 2.35 Angs. data (in fact, slightly better R-free). These values
> are likely to improve little bit after completing the solvent model. So, is
> this crystal really twinned?
>
> I have attached log files of cTruncate and most recent Refmac5 run with Twin
> refinement. Apologies for attachments (at least no image files).
>
> Thank you all in advance for educating me.
>
> -Partha
>

--
Ethan A Merritt
Biomolecular Structure Center, K-428 Health Sciences Bldg
University of Washington, Seattle 98195-7742