| Quick navigation: | Home | Site Map || References | Biography || Copyright | Other copyright | Contact us | | |
|
|
Re: [ccp4bb] Does NCS bias a randomly-chosen test set (even if not enforced)? |
|
CCP4bb navigationCCP4bb <-- 2008 <-- February 2008 <-- 09 February 2008Subject: Re: Does NCS bias a randomly-chosen test set (even if not enforced)? From: Edward Berry eaberry {- at -} LBL {- dot -} GOV Date: 2008-02-09 > >> (I'm probably wrong, but I want someone to show me,and not with >> hand-waving >> arguments or invocation of crystallographic intuition or such) >> >> To convince me, someone needs to show that the expected value of the >> change >> in Fo-Fc at a test reflection upon a change in the model (a step of >> refinement) >> is negative, even in the absence of any real improvement in the model, >> simply because the change reduces Fo-Fc at a sym-related working >> reflection. > > The problem is that a) the statistical drift will be very small, and b) If you can show me it is negative, I don't care how small. But if it would amount to a change of 0.001 in R-free I'm not going to worry too much about it! (Can you describe this "statistical drift" a little better?) > that this will be for *almost every* reflection in the test set. If it > were just a few, you'd be right, but not when it's all of them: then > your Rfree will not be informative. I don't get that- if the expected value is zero for each reflection, then the more you average the better it will approximate zero. > > That's in the absence of NCS restraints. In their presence, it's bad > anyway, because you're forcing Fc to be (almost) equal for both > reflections. For now I'm not arguing about that. > > phx. > > CCP4bb navigationCCP4bb <-- 2008 <-- February 2008 <-- 09 February 2008 |
| ProteinCrystallography.org: Copyright 2006-2008 by Quid United Ltd |