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

Re: [ccp4bb]: Jumping between refinement programs..



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

Tommi Kajander wrote:
> 
> ***  For details on how to be removed from this list visit the  ***
> ***    CCP4 home page http://www.dl.ac.uk/CCP/CCP4/main.html    ***
> 
> Dear all,
> Would anyone have suggestions about what is going on here:
> 
> I have bene refining my model with CNS, but as the unit cell is
> very large i thouhgt i might use ARP to add the rest of the waters
> --then i thought refine it a bit more with CNS, the R-factors
> at the at the beginning of the CNS conjugate gradient
> minimisation (with max. likelyhood target) very OK
> - better than i expected and better than what ARP/Refmac
> statistics gave me  --but the refinement  allways starts
> going upwards 'R-factorwise' in CNS
> --so the initial R-free is 2 % lower than the end value no matter
> what --Any suggestions for an explanations other than that its  a bad
> idea to jump between refinement programs..??
> I
> 've tried altering the ncs-restrians and the xray/geometry
> weighing.
> 
> I am just puzzled with the fact that the inital R-factors are allways
> the lowest!! Based only on R-factors (and geometry) one could take
> the model after ARP and say thats a done model -- its just that when
> one tries to refine it it tha doeasnt work. I have never had this
> kind of a problem with using diffrent refinement programs -- but it
> raises interestinn questions..
> 


 Are you preserving the same FreeR set in both procedures ? but if not
you are 
a) not using maximum likelihood properly, since the basis of it is that
there isa cross validation set to estimate the overall "error- ie D". 
and b) allowing Refmac/Arp/Warp to use your FreeR set for refinement,
which will obv. reduce the R factor for that set of data, then going
back into refinement alone, and excluding them. 

 The best way is to make sure that the mtz file and the CNS file share
the same Free R set - if you use f2mtz to make the mtz file with an
appropriate label for the test set; FreeR_flag is the standard, then the
uniqueify script it will sort this out .. 
uniquify -f FreeR_flag CNS.mtz 
 Eleanor Dodson