Dear all, We have tested the Charge Flipping algorithm with SUPERFLIP program on various experimental data (anomalous delta-F's, MAD FA's).
see http://www.cbs.cnrs.fr/SP/crystal/SUPERFLIP/ & Dumas, van der Lee, Acta Cryst D64, 864-73 In all successfull trials, a good quality substructure is obtained. As outlined below, the main requirements are good completion and high quality experimental data! This is a relatively new approach and there's still room for improvement. > Francis E Reyes wrote: > Orthorhombic C2221 using SUPERFLIP heavy atoms are found, ... > but the SG is wrong. Yes, this is the most probable diagnostic hypothesis. It looks like you have missed a space group. This ambiguity can arise with twinning, NCS in special position,... I had a similar case for Se substrcture solution using SUPERFLIP with data indexed as C2221, where P21 lattice approximates C2221. SUPERFLIP (http://superflip.fzu.cz/) can try to derive the symmetry operations of the P1 charge flipping density, independently of the symmetry entered in the input file (see Palatinus, J; applied cryst, 41,975-84) Restart a new job using the keyword derivesymmetry: the list of symmetry operations, centering vectors will be displayed (with agreement factors) in the .sflog file -'derivesymmetry no' the output density is in P1, no shifting or averaging or -'derivesymmetry use' shifting and averaging the output density. If you send me the inputs and sflog files I will have a closer look. Christian. -- Christian Dumas, Centre de Biochimie Structurale, 34090 Montpellier Cedex, France Tel: +33-(0)467.417.705 George M. Sheldrick wrote: > I have also played with charge flipping and my experience was the same as > Kevin's. Michael Woolfson once said that all 'direct' methods work fine with > perfect data. The method requires expansion of the data to P1 which seems > to degrade the quality of the solution; when the data are noisy, as is > usually the case with real anomalous differences, the space group can be a > useful constraint, as is the number of sites expected. If necessary one > can easily run the usual programs in all potential space groups. For the > solution of small molecule structures using atomic resolution (say 0.9A) > native data, the data are much less borderline and charge flipping in P1 > is a good way to explore phase space. > Another reason why charge flipping may not work so well with real > anomalous differences is that the data tend to be rather incomplete, > for example all the centric reflections are missing. This degrades > the quality of the resulting maps, which is more serious if you are > modifying low densities than when you are just searching for peaks. George Prof. George M. Sheldrick FRS Dept. Structural Chemistry, University of Goettingen, Tammannstr. 4, D37077 Goettingen, Germany Tel. +49-551-39-3021 or -3068 Fax. +49-551-39-22582 Kewin Cowtan wrote: > I played with this (coded from scratch, both simple algorithm and a few > tweaks) for a couple of weeks for solving heavy atom substructures. With > perfect FAs it works well and quickly. With real delta-F's it didn't > work at all. Can't remember if I tried perfect delta-F's. > Probably SUPERFLIP is better than my quick implementation though. Francis E Reyes wrote: > Hi all > I've been playing around with charge flipping for macromolecular > substructure determination with pretty promising results. I'm > particularly attracted to the fact that it solves structures in P1, > with no space group assumptions and curious how it would handle some > of the pseudosymmetry cases I've come into in my time. > > I'd like to know if anyone's had experience with this method, and > open up the discussion with the following questions: > > As the algorithm starts with completely random phases and charge > flips the map in P1, what is the importance of measuring (good or > any) anomalous signal at all (for the sole purpose of finding the > heavy atoms)? At first pass it would seem that just as long as you > have an incorporated heavy atom and the density of that region is > greater than delta, that this alone would be sufficient for locating > the position of the heavy atom. In other words just as long as your > heavy atom is sufficiently higher in contrast than your protein/rna it > would be a good enough criteria. > > In the above regime, would the importance of measuring anomalous data > be more important for substructure refinement (via phaser, mlphare, > sharp, solve/resolve)? > > Now to a more specific question for those who've had experience (or > maybe the authors are subscribed here): > > Orthorhombic C2221 using SUPERFLIP heavy atoms are found with great > peakiness (before noise suppression: peakiness = 5, after noise > suppression peakiness >25, good separation of heavy atom peaks from > noise peaks in resulting pdb). Yet the space group check via the sym > operators is rather poor (overall agreement close to 100). My > interpretation is that the heavy atoms are found, but the space group > is wrong? > > Thanks! > > --------------------------------------------- > Francis Reyes M.Sc. > 215 UCB > University of Colorado at Boulder >