This reminds me of an argument made by James Holton in a previous post, that
to a first approximation it is the number of photons collected for each unique reflection that determines the quality of the data, and this can be done using a larger number of frames with shorter exposure, or a smaller number of frames with longer exposure. With shorter exposures I/sigma for individual measurements goes down, but this is made up for by the greater number of measurements being averaged, so Rpim and I/sigma for averaged individual reflections will be about the same. CC1/2 also should be about the same, because the data for each reflection are split in half and the averages of the halves are compared, those averages will be more accurate if they involve a larger number of reflections. HOWEVER Rmeas (or Rmerge) measure the reliability of individual measurements so they do not benefit from redundancy and will become very bad if the exposure is short.
In general R-meas is a better criterion than Rmerge, but I think they will be almost the same for redundancy 9 or greater.
I wouldn't exclude 80% of the frames unless there is clear evidence of radiation damage. Does the merging B-factor go up in the series by greater than 10? Do other statistics like CC1/2 and Rpim improve on excluding the reflections?
I think part of Holton's point was that if you go for short exposures and many frames, then if the crystal turns out to be super-sensitive and dies early, you may still have a complete albeit noisy data set.
eab
On 07/20/2016 11:07 AM, Gergő Gógl wrote:
> Dear all,
>
> I am trying to process a weak low resolution data which was crystallized and collected in an other lab but unfortunately with suboptimal crystal handling (cryo...) and data collection strategy (1° oscillation, close detector distance...). The data is highly redundant but the Rmeas is really bad. We already suggested them to collect better data from a better crystal but it seems to be difficult for them...
>
> The overall data has a redundancy of 40 (43 in the highest bin) with an overall Rmerge 75% (365% in the highest bin) while the overall CC1/2 is 99% (83% in the highest bin). (The XSCALE.LP is attached for the whole dataset.) I was able to decrease the overall Rmerge to 36% by discarding ~80% of the collected frames but it is still a marginal data (with a redundancy ~9). On the other hand the refinement gave us a reasonable structure with good Rfactors (Rwork 22% Rfree 26%). (It is a protein-peptide complex where we are interested in the bound state of the peptide.)
>
> We are in a disagreement in our lab and already asked a few crystallographer but did no reached a clear consensus answer. Is this data acceptable to publication? Can you trust a data like this?
>
> Best,
> Gergo Gogl
|