Engin Özkan wrote:
> On 7/22/13 11:20 PM, Ethan Merritt wrote:
>>
>> Which brings up the point that something seems to have
>> gone wrong in one of your processing runs.
>> Both runs claim mean (I/sigI) in the outer shell is 2.0,
>> but in one case this is for the 2.4A shell and in the other case
>> it's for the 2.2A shell. That is unlikely to be correct.
>> I/sigI should not depend on the Laue group.
>>
> Why is this surprising? This is what I would expect. By going to
> orthorhombic, she is getting higher redundancy (two-fold more), and
> hence higher I/sigI. She will also have fewer molecules per asu, and
> less of NCS to help her with refinement if she sticks to orthorhombic.
>
Here we need some clarification on which I/s(I) is meant - sigma(I) for the
individual measurements, or sigma(I) from error propagation to the final
<I> which as Engin notes will be lower.
I thought it was the former that is to be used for selecting the cutoff, and
this is somewhat confirmed by the the recent Aimless paper (ActaD 69 1204-1214
"How good are my data and what is the resolution?" Philip R. Evans and Garib N. Murshudov):
"The `maximum resolution' is estimated from the point at which <I/[sigma](I)> falls below 1.0 for each batch: note
that this <I/[sigma](I)> is without averaging multiple measurements (which would not generally occur on the same image),
so will be smaller than the <I/[sigma]> after averaging."
But I may be misunderstanding the point of that sentence.
Also note sigma(I) depends on your error model (at least for scalepack addicts)
> But, other than that, yes, twin R factors should come with big warning
> signs on them. Intensity moments, for example,
>
> <I**2>/<I>**2 = 2.032
>
> are showing no signs of twinning (and there is no translational NCS to
> confuse the intensity stats), so the data should not have been treated
> as twinned in the first place. I agree with Ethan that at first sight,
> it appears orthorhombic.
>
> Engin
>
|