I should make the estimation in Aimless more robust, and curve fitting sounds like a good idea (but what function?). Outliers are a difficult problem, but anyway I think you should look at the curve and not just the number estimated. I would look at I/sigI as well, and anisotropy to decide the resolution. However, the final cutoff should probably be based on refinement, and also I don't think the exact cutoff makes a huge difference (see http://www.ncbi.nlm.nih.gov/pubmed/23793146)
Phil On 15 Aug 2014, at 15:54, Ed Pozharski <[email protected]> wrote: > Same here. Ultimately, the KD test must be used in the end to finalize the > resolution (keeping in mind recently discussed issues of effective resolution > given data completeness). I just want to add that at least some versions of > aimless report overestimated resolution based on CC1/2 cutoff when outliers > are present (e.g. due to ice rings or salt diffraction). It seems that > aimless just picks the highest resolution bin where cc1/2> 0.5 even if some > lower resolution bins are below 0.5 as well. I have written a script for more > robust automated evaluation of these curves. In a nutshell, it fits CC1/2 > (d) curve to 1/(1+exp (-x)) and returns the resolution at midpoint. I'm > pretty sure that theoretical CC1/2 (d) dependence is different from this, but > it seems good enough for a rough estimate. > > > Sent on a Sprint Samsung Galaxy S® III > > > > > > > > > -------- Original message -------- > From: Roger Rowlett > Date:08/14/2014 5:44 PM (GMT-05:00) > To: [email protected] > Subject: Re: [ccp4bb] CC-half value ?? > > Exactly. Aimless will give you suggested resolution cutoffs based on CC 1/2 > in the log file. > > Roger Rowlett > > On Aug 14, 2014 5:04 PM, "conan仙人指路" <[email protected]> wrote: > Hi Faisal, > > CC-half standard is valuable in evaluating the cut-off of highest > resolution. Sometimes even if I/sigI is close to 1 and completeness is not as > high, if CC-half is still significant, it may be worth incorporate the extra > high-res shell data and extend the resolution. Again, if only the reliability > and unbias are carefully confirmed, and the apparent significant CC-half is > not due to an artifact of some other factors like ice ring etc. > (Ref: Karplus PA and Diederichs K. 2012 Science 336, 1030-1033 > https://www.pubmed.com/pubmed/22628654) > > It has yet to be appreciated by most population of the crystallography > society, unlike the I/sigI, completeness, Rsym. In particular, Rsym has > gradually less a direct measurement of the data quality and or determinant of > resolution cut-off. > > Best, > Conan > > Hongnan Cao, Ph.D. > Department of Biochemistry > Rice University > > Date: Fri, 15 Aug 2014 01:39:48 +0530 > From: [email protected] > Subject: [ccp4bb] CC-half value ?? > To: [email protected] > > Dear all > > How CC-half value of a data set determines the maximum resolution limit > during data processing ?? Although much we know about the Rsym and I/Isig > values of the highest resolution shell while processing the data, what are > the parameters we need to check related to CC-half values ?? > > -- > Regards > > Faisal > School of Life Sciences > JNU >
