 |
 |
Thanks all - also for jogging my memory. Peter, I think Garib ended up doing something statistically sophisticated, but I couldn't track down the slides or documentation I remember seeing somewhere. But I didn't dig very hard, and if it's been there since 2012, then that explains why I failed. Thank god your supervisor and his/her ilk haven't had sole discretion over the research funding that's been supporting the algorithm and method developers all these decades; if they had, we'd still in the year 2020 be wasting millions per project just to get those 200um crystals that yield the signal needed for the ancient algorithms to recognise the presence of a diffraction spot. The other quiet revolution in structural biology: algorithms. Frank On 29/04/2020 21:53, Petrus Zwart wrote: > I am not quite sure how it is done in Refmac, but in phenix it works > as follows > > - difference maps are gradient maps for a least squares target and > they do not contain any amplitude contamination from twinning. > > - Fo type maps are computed either via classic detwinning (solving the > linear equations) or this is done using some proportionality rule like > James and Eleanor allude to. > > An alternative way would be use something like this relation > > Map coefficients we want: 2Fo-Fc > > Realize that 2Fo - Fc = Fc + 2(Fo-Fc) > > and compute > > Fcalc + 2*difference map*magic scale factor > > It is a bit of a hack, no FOM's etc and I worry about bias. It would > be far better to get map coefficients from the joint distribution of > the structure factors that take into account twinning. In this way, > you can handle experimental error as well in a proper fashion. I > wouldn't be surprised if this happens in Refmac, it doesn't in Phenix > (yet). > > My PhD supervisor's advice always holds btw: "Grow better crystals" > > Regards > Peter > > > > > > On Wed, Apr 29, 2020 at 12:40 PM James Holton <[log in to unmask] > <mailto:[log in to unmask]>> wrote: > > Yes, they are "de-twinned". There is really no other way to do it. > Without de-twinning you'd be looking at a map of two overlapping > structures. How do you turn a single observation of the sum of > intensities from two different hkls into two different Iobs > values? You > need to know the ratio. The twin fraction is one way, but once it > gets > close to 0.5 the "de-twinning" gets very noisy. I believe what the > refinement programs do is use the ratio of the two Fcalc^2 values in > order to "split" the Iobs value. Doesn't that introduce "bias"? > Yes, of > course it does. That's one of the many reasons why twinning sucks. > > Now, of course, there are probably fancy weighting schemes that make > things a bit better, but rather than speculate I'd ask the program > authors themselves. The best thing, of course, is to refine until > Icalc > becomes more accurate than Iobs, which is the typical result of > small-molecule refinements. I expect that is why they are not > nearly as > afraid of twinning as we are. > > -James Holton > MAD Scientist > > > On 4/28/2020 2:19 PM, Frank von Delft wrote: > > Hi all - feel free to point me to the docs if it's already clear > > somewhere: > > > > When refmac generates 2mFo-DFc maps after twinned refinement, > are they > > the "untwinned" view of the electron density? I.e. with the > twinnig > > convoluted out by his statistical magic? > > > > I remember Garib mentioning the ambition a few years ago; but I've > > not been paying enough attention to remember if it's actually > > implemented, or still in progress (because it's hard). > > > > > > Thanks > > Frank > > > > > ######################################################################## > > > > To unsubscribe from the CCP4BB list, click the following link: > > https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > > ######################################################################## > > To unsubscribe from the CCP4BB list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > > > > -- > ------------------------------------------------------------------------ > P.H. Zwart > Staff Scientist > Molecular Biophysics and Integrated Bioimaging & > Center for Advanced Mathematics for Energy Research Applications > Lawrence Berkeley National Laboratories > 1 Cyclotron Road, Berkeley, CA-94703, USA > Cell: 510 289 9246 > > PHENIX: http://www.phenix-online.org <http://www.phenix-online.org/> > CAMERA: http://camera.lbl.gov/ > ------------------------------------------------------------------------- > > ------------------------------------------------------------------------ > > To unsubscribe from the CCP4BB list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > ######################################################################## To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1