[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [ccp4bb]: map coefficients in Refmac and CNS



***  For details on how to be removed from this list visit the  ***
***          CCP4 home page http://www.ccp4.ac.uk         ***

> > Yes, but the matter is that such difference densities are quite
> "different" 
> > (one from the other) as output by refmac and CNS, which I was thinkig
> were due 
> > to refmac "combined" phases PHWT and PHDELWT. So I would like to know
> whether 
> > it is always better to calculate my weighted 2Fo-Fc and Fo-Fc maps
> with them or 
> > in some situations I should use PHIC as I am supposing CNS always does
> ...
> >   Anyhow, there seems to be a good sight as the Fourrier difference in
> CNS show 
> > the "whole ligand". But why should REFMAC later on break its density ?
> For 
> > curiosity, I also fft'ed a map with FWT and PHIC (output from a Refmac
> run 
> > where I input also the ligand) and the ligand looks much better than
> with FWT 
> > and PHWT ...
> 
> Oh, but that should be disturbing....  how long have you been refining
> with the ligand?  Did you build the ligand right from the start, and

  After several cycles in CNS, then the density showed the ligand, then it was 
put there and was (somehow) happy within CNS, then I moved to REFMAC and the 
problems started ... I was wondering if if I could attribute this difference to 
PHIC/PHIWT...

> have been refining away for 20 rebuild cycles until somebody asked to
> see difference density?  In that case, the image of the ligand gets

  After several cycles in CNS it was put and after this several others in CNS. 
As said above, the problem was nticed at moving to REFMAC and calculating the 
maps recomended by it.

> "imprinted" in the rest of the model (the errors, i.e. the b-factors,
> mainly) and it reappears in difference maps.  Basically a version of
> model bias:  the "wrong" model (the potential ligand) is accommodated by
> adjusting the errors.
> 

  Thanks for the warning ! In fact, in the above cited test (Fourrier 
difference) I simulated annealed (5000 K, 25 K step - is this enough ?) for 
taking out bias fom the model, but in CNS the ligand showed up in the way I 
described before. Only in CNS.

> Ways to remove that would be to shake the coordinates with random
> displacements (e.g. the NOISE keyword in pdbset), reset the b-factor,

 I suppose I made that with the sa.

> rebuild it from scratch (arp/warp is handy), or start from the
> beginning, but leaving the ligand (and other dodgy bits) till the very
> last.  This is actually a general principle I follow for building

  I am also doing that. But still obsessed with the difference in CNS and 
REFMAC. Should not REFMAC coefficients lead to a better map ? If its map is 
better, my ligand is not properly there, and CNS must bias the maps a great 
amount at using phic .... Although, might there be cases where refmac maps are 
worse than the traditional ones calculated with phicalc ?

> structures.  
> 
> I don't honestly know what the differences in refmac and CNS maps are
> (if any), but I understood the weighting's the same. 
> 

  You man, the weighting on the coefficients ? Yes, now it seems to me that m 
and D are calculated based on the free set in both programs (not aware though 
how CNS deals the absent reflections), but the phases I still do not know.

> Good luck,
> 
> phx.
> 

  Thanks once again,

Jorge.