Answering the question "should I even bother trying?" can be
complicated, but I get asked that a lot at the beamline! I recently
incorporated a number of data quality formulas into a little interactive
web page here:
http://bl831.als.lbl.gov/xtalsize.html
which focuses on calculating how many crystals of a given size you will
need to average together to attain a specified goal (either weak spots
(high res) or subtle differences (anomalous)) given radiation damage
limits.
For those interested in explanation:
A quick-and-dirty way to estimate your Bijvoet ratio is with the formula:
0.75*fpp*sqrt(sites/MW)
where "sites" is the number of sites you expect per "MW", where "MW" is
in Daltons (amu). Note that "MW" can represent the protein monomer,
asymmetric unit, or whatever, as long as "sites" refers to the same
thing. "fpp" is the expected number of "anomalous electrons", which you
can get from the CCP4 program "crossec" or several websites.
Now, since you didn't mention your molecular weight, I will guess that
it is about 30 kDa, which means your Bijvoet ratio at CuKa (where the
fpp of sulfur is ~0.56 electron) will be about 0.6%.
Now, detecting a 0.6% difference requires that the two things you are
subtracting (I+ and I-) be measured to at least a precision of 0.42%
(because the error in the difference will then be: sqrt(0.42%^2+0.42%^2)
= 0.6%). This is the BARE minimum (where the signal is equal to the
noise), but even to reach this goal, your signal-to-noise ratio must be
1/0.42% = 240. This is challenging! Typical data sets have I/sig(I) ~
30 in their "best" bins (see Deiderichs 2010
http://dx.doi.org/10.1107/S0907444910014836). A multiplicity of 23 can
push I/sig(I) to 30*sqrt(23) = 143, which is good, but still not close
to 240. You will probably need a multiplicity of ~65 to measure Bijvoet
differences to an accuracy of 0.6%
Then again, if your protein is only 10 kDa, then your Bijvoet ratio is
~1% and you will only need I/sig(I) > 140.
-James Holton
MAD Scientist
Vandu Murugan wrote:
Dear all,
I have collected a 2.7 angstrom home source data with Cu-Kalpha
source for a protein with 6 cysteines, with a multiplicity of around
23. I need to know, is there any significant anamolous signal present
in the data set, since there is no good model for my protein. Can any
one tell, which program to run, and what parameter to see? Thanks in
advance.
cheers,
Murugan
