Hi ,
i just tried the extra information because I had a serious "overfitting" of my
model. Roughly 10 % difference for 1.9 Ang. resolution and experimantal
phasing. I thought maybe the extra information and the adjustment of the
weighting parameters helps me to figure this out. A refinement using the
experimental phases brings indeeed R values to a more reasonable difference
(5 %). To be honest. I adjusted also the weight manually to a more stringent
value (0,1 weighting term). I am not sure what of the both adjustments was the
more important one, but after an adjustment of the model I could probably do
the rest of the refinement without the additional phase information.
For the record the my sharp Arp/Warp had already a difference of roughly 10 %
and had already some overfitting.
It was a bit strange for me but I thought it will be resolve during the
manual model adjustment and refinement.
I probably made a lot of wrong decisions in the beginning, but I have learned
a lot at the end for the next structures; I hope at least.
Thanks for for this interesting disussion.
Christian
Am Montag 11 Oktober 2010 22:31:24 schrieb Tim Gruene:
> Hello Pavel,
>
> thank you for the PDF file. When I search for the word "phase" starting
> from page 24, all I find are "phenix.phase_and_build" and
> "phase_and_build_params.eff" on p. 34, and they don't explain the problem.
>
> I agree that experimental phases are data, but they might be utterly noisy
> data and therefore hinderous rather than helpful.
>
> I can easily integrate a data set to 1.0A and thus have data to 1.0A, if I
> set the detector distance close enough, but if my crystal only diffracts
> to 2A, the data between 2A and 1A only will do harm to the map. That's my
> main point, not the meaning of TLS
>
> Cheers, Tim
>
>
>
> On Mon, Oct 11, 2010 at 12:52:03PM -0700, Pavel
>
> Afonine wrote:
> > Hi Tim,
> >
> >> following this discussion, I was wondering how much sense it makes to
> >> use TLS domains and experimental phases at the same time.
> >
> > I guess both things are completely irrelevant. With using experimental
> > phases you bring in more data. And with using TLS you simply use a more
> > adequate model for ADPs. It's explained here:
> >
> > http://www.phenix-online.org/newsletter/CCN_2010_07.pdf
> >
> > starting page #24.
> >
> > All the best!
> > Pavel.
>
|