[phenixbb] Table 1 stastics

[Kip Guja]

Hi Nat and Bjørn,

I apologize if this is way off base, but perhaps the difference in completeness is due to the (potential) difference between I-obs / F-obs (original reflections after data processing and reduction) and I-obs-filtered / F-obs-filtered (reflections actually used in refinement).  

Specifically, I am referring to data that was refined in older versions of Phenix (before French&Wison was default) -- and datasets that have some amount of weak reflections or negative intensities. Perhaps that is the case here?

An easy to tell if there is a difference is to open an output mtz from refinement in ViewHKL (comes with CCP4 6.3) -- it nicely lists the original data vs data used for refinement and the corresponding completeness for each.

Hope that helps,

Kip

On Oct 25, 2012, at 1:57 PM, Nathaniel Echols wrote:

> On Thu, Oct 25, 2012 at 9:59 AM, Bjørn Pedersen <bjopp at msg.ucsf.edu> wrote:
>> I was under the impression that the differences arise due to XDS uses
>> SIGNAL/NOISE >= -3.0 as the cutoff, thus allowing some negative reflections,
>> while phenix use a strict cutoff of 0? Note, that the completeness in the
>> high res. shell is lower in phenix than XDS (90.6 vs 99.7), a behavior I
>> have seem several times solving different structures. XDS will show high
>> completeness at high res. and then in phenix.refine it always drops.
> 
> Phenix used to have a strict cutoff, but it now runs French&Wilson
> treatment by default, so negative intensities are allowed.  However:
> Kay pointed out that in another piece of code, I am calculating
> I/sigma starting from amplitudes, which are always positive.  For
> Table 1, if you use intensities as input, this won't matter, but if
> you use amplitudes and many of the original intensities were negative,
> it may be overestimating I/sigma.  Fixing this now.
> 
> It still doesn't explain why the completeness would be so far off, however.
> 
> -Nat
> _______________________________________________
> phenixbb mailing list
> phenixbb at phenix-online.org
> http://phenix-online.org/mailman/listinfo/phenixbb