Hi -
Its not clear to me if the 2.3 A dataset is from a different crystal,
if it has the same space group as the 2.7 A dataset and if it is
isomorphous, and how the 'jump' from the 2.7 A dataset to the 2.3 A
dataset has been done. I can think of quite a few ways to do that
transition, and each way can carry its own problems.
Maybe its worth clarifying these in more detail.
I also have to say that (for once!) I dont agree with Kay. I have
tried in the past to refine against datasets processed with different
programs, and I know people that do that a lot; the differences are
minor at best (1-2 % Rfree, mostly differences are in the anomalous
signal quality). Thus I dont think its processing but a mistake in
the settings or somewhere else something that went clearly wrong
(different origin ?). Anyway, i am not sure about that either, since
Satinder says that the model does fit the density (which density
though ? Also not clear).
BTW, assuming that the two datasets are from different non-isomorpous
crystals, thats what I would be tempted to try and do:
1. Use 'phaser' to do a molecular replacement of the 70% complete
model to the 2.3 A dataset
2. Use 'cad' to combine the output mtz file of 'phaser' with phases
and weights, with you SAD phases and weights, plus Fobs of both
datasets.
3. Use dmmulti to do 'multi-crystal averaging' using the molecular
replacement phases of the 2.3 A dataset and the 2.7 A SAD phases.
Also use the NCS. That must result to a good 2.3 A map.
4. Use ARP/wARP to autotrace the 2.3 A map with the 2.3 A fobs.
I should also add that if I had a 2.3 A native and a 2.7 A SAD map in
NON-isomorphous crystals in the first place, I would use DM-multi
directly,
even without phases for the 2.3 A map. That has worked beautifully
for me in the past. Anyway - too late for that now since you have a
model.
Last but not least if the 2.3 A dataset and the 2.7 A dataset are
isomorphous, I would most definitely have tried phasing using both of
them
Especially if one had eg Se and one did not (the best SAD experiment
will give you 6 to 7 e to use; Se-S has 18 e. why not use them if you
can ??
yes, SAD might sometimes be the best map as many people have shown
over the last few years,
but also sometimes the Se-derivative - Native has lots and lots of
info and using all info can give better maps when non-isomorphism is
not a problem.)
May I also assume that when you are sure of the space group that
means you put it through x-triage and its not a disguised P21 twin,
as many P21212 cases I have seen.
Tassos
On Jul 27, 2007, at 2:19, Satinder K. Singh wrote:
Hello,
I have a SAD data set to 2.9A and a native data set on the same
protein to 2.3A. I have built 70% of the model (polyalanine) into
the SAD experimental map that has been subjected to 2-fold NCS
averaging and phase-extended to 2.3 A. From the maps, the model
looks like it fits the density fine, but when I try to refine
(either rigid-body or positional refinement in REFMAC or simulated
annealing in CNS), I get an R-factor of ~60%, worse than random.
Similarly, a SigmaA run ("combine isomorphous phase with partial
structure") on the model yields an R-factor of ~60%. However, when
I run DM on the original phase-combined file (before NCS averaging
and phase extension) in omit mode, I get an R-free of 34.6%, which
suggests that the map files itself is okay.
I also checked the space group of the processed data, and it is
definitely P21212.
Does anyone have any suggestions of what I may be doing wrong?
Thanks in advance for your help.
Kind regards,
Satinder
