see below (1) Indeed, < <I>/σ(<I>) >, aka Mn(I)/sd(Mn(I)) is as quoted by Clemens. Is this confusing? I thought we had been very clear. The value reported in Scala and Aimless as "Mn(I/sd)" have the same definition, and I believe it's the same in XDS. Note that the sigma(I) values will depend on the SD "corrections" Sdfac, SdB, SdAdd and they can sometimes go a bit crazy (I'm working on this). This is one reason why CC(1/2) may be more useful
Obviously including extra measurements in the average (increasing multiplicity) will improve the data, unless there is something wrong with the added data, notably radiation damage
see below (2) The other quote from section 3.1 of our paper is an analysis against "batch" i.e. image number or time, where a plot of I/sig(I) is done for each batch, and the resolution at which this signal/noise estimate falls below (arbitrarily) 1.0 is plotted against batch number. This is done before averaging as there are unlikely to be a significant number of symmetry equivalents on any one image. This plot is _not_ intended to give a true resolution estimate, but is there to show trends in the strength of the data (signal/noise) through the data collection, with effective resolution maybe varying due to such things as anisotropy, illuminated volume (e.g. plate crystals) or radiation damage. The plot may be one piece of information in deciding whether to cut e.g. the end of the data, in conjunction with the cumulative completeness plot
Phil
On 23 Jul 2013, at 16:29, Clemens Vonrhein <[log in to unmask]> wrote:
> Hi,
>
> On Tue, Jul 23, 2013 at 10:11:04AM -0500, Engin Ozkan wrote:
>> Isn't the reported Mean(I/sigI) in the reported Aimless table for
>> the merged/averaged reflections (because that is what we are
>> discussing about)?
>
> Yes - according to
>
> 1. http://ccp4wiki.org/~ccp4wiki/wiki/index.php?title=Symmetry%2C_Scale%2C_Merge#Analysis_by_resolution
>
> "... the average signal/noise after averaging symmetry-related
> observations < <I>/σ(<I>) >, labelled Mn(I)/sd(Mn(I)) in the
> Aimless table, ..."
>
> 2. section 3.2.1 of http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3689523/
>
> "... from the average signal-to-noise ratio of the merged
> intensities as a function of resolution. ... the average intensity
> over symmetry mates <I h> is divided by its estimated error σ(<I>)
> and this ratio is averaged in resolution bins [reported as Mn(I/sd)
> in the program output].
>
> So the average (denoted by < and > in the output) is over all unique
> reflections within a resolution bin - after all, those stats are
> reported within resolution bins. And each unique reflection is the
> result of (weighted) merging of all its measurements.
>
> ... as far as I understand this ...
>
> Cheers
>
> Clemens
(2)
>> 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."
>>
> Could it be that the reported <I/sigI> an average of the batch-wise I/sigI's? I would love to hear that confirmed (or denied) by the authors of Aimless.
|