I hope everyone understands that I am not the corresponding author on
this paper! That is Henry Chapman. He, and most of the other authors
have been slaving away on this problem for most of their professional
careers. To be honest, I never did think it was going to work, and I
was absolutely "gobsmacked" as they say when I heard they got spots
going all the way out to the edge of the detector!
However, since my involvement was mostly about the data processing and
refinement, I think Henry won't mind if I respond to a few of the
questions posted on the BB about it.
Hmm. Looks like our paper didn't report the R/Rfree for the "control"
data set. Crap. Sorry about that. For the "record", it was
0.27/0.30. These "control" data were collected from a large and
cryo-cooled photosystem I crystal at ALS 8.2.2, truncated to 8.5 A and
artificially "twinned" by averaging F(h,k,l)^2 and F(k,h,-l)^2, then
subjected to the same rigid-body REFMAC refinement as the LCLS data.
Exactly the same Rfree flags were used for both cases, and these were
chosen in the point group 622 so that no "twin mates" were in both the
working and test sets.
And before you ask, no, I have no idea why the "control" data were
"worse". However, I can say that they were not all that different from
the LCLS data. The maps (which were shown) don't really look all that
different (IMHO), and the R-iso between the ALS and LCLS datasets was
22% (also shown). I thought the latter was amazing agreement
considering that the ALS crystal was cryo-cooled, the LCLS crystals were
at room temperature and the lowest R-iso between lysozyme datasets in
the PDB collected at room temp vs cryo is 20% (2hu1 vs 2epe). Yes, I
checked all 10,000 lysozyme-lysozyme pairings.
I'm afraid I can't directly address Hudel's Rtwin = 0.7*Rnormal rule
because de-twinned LCLS data are not yet available and I don't have the
untwinned ALS data on my computers. The reason I was "impressed" was
because data in this resolution range generally don't fit very well to
molecular models (have a look at R/Rfree in the infinity-9A bin of your
last structure). So, I was surprised to see that R/Rfree was "okay". I
guess I am just a pessimist. All I can really say is that we used
Refmac_5.6.0076, and the data are deposited under 3pcq if anyone wants
to play with it.
On 2/9/2011 10:28 PM, Hudel Luecke wrote:
> I must admit not (yet) having read the "data processing paper PMID: 20389587" nor the 88 author paper. Nevertheless, I would like to comment on your remark "Personally, I was quite impressed by how good the R factors were, all things considered."
> If I am not mistaken, perfectly twinned intensity data, such as you seem to have been dealing with, will generate data sets with compressed dynamic ranges, which in turn means that crystallographic R factors computed based on such data sets will be lower by roughly sqrt(2) or 1.4 compared to a non-twinned data set. So a somewhat iffy R(cryst) of 30% would look quite nice (21.4%) when computed on a hemihedrally twinned data set. But maybe this is all discussed in your paper...
> Cheers, Hudel