Alun,
Kay is right on target; I think there is no debate about that you must
use one FreeR set for all (even marginally*) isomorphous crystals. Your
colleagues who are not following this practice are not doing themselves
any favors.
If you don't keep the same FreeR set, R & Rfree for that 2nd crystal
will be almost identical --- and low!, like the R from crystal 1. Best
practice is to create an initial, master FreeR set that extends well
beyond your current (first crystal) resolution...crystals usually get
better as time spent on a project goes on, and extending the set
correctly takes some careful neuronal work (admittedly, now a bit easier
with the CCP4i GUI).
Dave
* I've not seen any arguments for why one *shouldn't* keep the same set,
even as isomorphism fades away, due, e.g., to a slightly variable unit
cell. Keeping the same set does no harm; calculating the potential harm
due to switching sets seems not worth the bother (though I guess one of
our more theoretically-inclined contributors will see an opportunity
there!).
> -----Original Message-----
> From: CCP4 bulletin board [mailto:[log in to unmask]] On
> Behalf Of Kay Diederichs
> Sent: Thursday, December 18, 2008 3:05 PM
> To: [log in to unmask]
> Subject: Re: [ccp4bb] Transferring a Free R set.
>
> Alun R. Coker schrieb:
> > Hi All,
> >
> > I have been in the habit of transferring my initial free R
> assignments
> > to any new data sets or to isomorphous data sets such as substrate
> > complexes. Although theoretically this is necessary to
> obtain a valid
> > free R many of my colleagues maintain that this is completely
> > unnecessary in practice. Does anyone on the list have a
> view on this or
> > has anyone tested to see if it makes any difference.
> >
> > Alun.
> >
>
> Alun,
>
> I completely agree with you about the way how to treat R-free
> reflections. If you want to have an unbiased R-free then you
> need to set
> these reflections aside for all refinement calculations of a project,
> and this applies to all datasets you collect, as long as they are
> isomorphous.
> This is especially important for low resolution work where
> the danger of
> overfitting is highest.
>
> Kay
>
|